40 results about "Feed forward neural" patented technology

A color management method / apparatus generates image color matching and International Color Consortium (ICC) color printer profiles using a reduced number of color patch measurements. Color printer characterization, and the generation of ICC profiles usually require a large number of measured data points or color patches and complex interpolation techniques. This invention provides an optimization method / apparatus for performing LAB to CMYK color space conversion, gamut mapping, and gray component replacement. A gamut trained network architecture performs LAB to CMYK color space conversion to generate a color profile lookup table for a color printer, or alternatively, to directly control the color printer in accordance with the a plurality of color patches that accurately. represent the gamut of the color printer. More specifically, a feed forward neural network is trained using an ANSI / IT-8 basic data set consisting of 182 data points or color patches, or using a lesser number of data points such as 150 or 101 data points when redundant data points within linear regions of the 182 data point set are removed. A 5-to-7 neuron neural network architecture is preferred to perform the LAB to CMYK color space conversion as the profile lookup table is built, or as the printer is directly controlled. For each CMYK signal, an ink optimization criteria is applied, to thereby control ink parameters such as the total quantity of ink in each CMYK ink printed pixel, and / or to control the total quantity of black ink in each CMYK ink printed pixel.

The invention discloses a method and a system for detecting voice spoofing attack of speakers on the basis of deep learning. The method includes constructing audio-frequency training sets, initializing deep feed-forward neural networks and deep recurrent neural networks and respectively training the deep feed-forward neural networks and the deep recurrent neural networks by the aid of multi-frame feature vectors and single-frame vector sequences of the training sets; respectively leading frame level and sequence level feature vectors of to-be-tested audio frequencies into two trained linear differential analysis models in test phases, weighting two obtained result grades to obtain scores and comparing the scores to predefined threshold values so as to discriminate voice spoofing. The method and the system have the advantages that local features can be captured, and global information can be grasped; the linear differential analysis models are used as classifiers in identification and verification phases, the voice spoofing attack can be judged by means of grade fusion, and accordingly the voice spoofing detection accuracy can be greatly improved.

The invention discloses a feature loss-based medical image super-resolution reconstruction method. The method involves an image conversion network fw; the image conversion network fw is a feed-forward full connection neural network; neurons in the feed-forward full connection neural network are divided into different groups according to an information receiving sequence; each group is regarded as a network layer; the neurons in each layer receive numerical outputs of the neurons in the previous layer to serve as inputs of themselves, and outputs of themselves are input to the next layer; information in the whole network is transmitted in a direction; and the image conversion network fw receives a low-resolution medical image having a size of H / 4XW / 4 and sent by the feed-forward neural network, and converts the low-resolution medical image having the size of H / 4XW / 4 into a high-resolution image having a size of HXW.

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 invention discloses a performance prediction method applicable to dynamic scheduling for a semiconductor production line. An extreme learning machine (ELM) is applied to prediction and modeling in the performance prediction method. Feeding control and scheduling rules are considered in a unified manner in the method, short-term scheduling key performance indexes such as an equipment utilization rate and a movement step number are predicted on the basis of a real-time state of a system, and a foundation is provided for dynamic real-time scheduling. A novel feed-forward neural network of the ELM is introduced into the semiconductor manufacturing system, and a prediction model is built by the aid of available data of the production line. As shown by test results, ideal prediction results can be quickly acquired by the method implemented by the aid of the ELM, the method has obvious advantages and an obvious application prospect in the aspects of parameter selection and learning speed as compared with the traditional neural network modeling method, and a new idea is provided for online optimal control.

The invention relates to an elevator risk evaluation method, and belongs to the technical field of elevator risk real-time evaluation and process and safety control. The elevator risk evaluation method comprises the steps that (1) an index analysis method and an analytic hierarchy process are used for analyzing factors which possibly cause elevator faults, and an elevator risk evaluation index hierarchical structure is built; (2) actually-measured parameter values of indexes of the elevator risk evaluation index hierarchical structure are obtained, and initialization processing is carried out on the actually-measured parameter values of the indexes; (3) the weight of the indexes of the elevator risk evaluation index hierarchical structure is determined through an artificial neural network; (4) risk evaluation is carried out on an elevator through a feed-forward neural network with a time sequence learning algorithm, and the elevator risk grades are divided. The elevator risk evaluation method has reference value and application value in a plurality of fields, and can be applied to risk evaluation of industrial boilers, large-sized power transformers, industrial pipelines and the like.

The present invention is directed to a deep neural network (DNN) having a triplet network architecture, which is suitable to perform speaker recognition. In particular, the DNN includes three feed-forward neural networks, which are trained according to a batch process utilizing a cohort set of negative training samples. After each batch of training samples is processed, the DNN may be trained according to a loss function, e.g., utilizing a cosine measure of similarity between respective samples, along with positive and negative margins, to provide a robust representation of voiceprints.

The invention discloses a cardiovascular disease (CVD) prediction model using an extreme learning machine (ELM) algorithm based on single-hidden layer feed-forward neural networks (SLFNs), which means an ELPSO-ELM hybrid intelligent algorithm prediction model. The ELPSO-ELM hybrid intelligent algorithm prediction model is designed by using the algorithm through the following steps: building a leadership particle enhanced particle swarm optimization algorithm (ELPSO) by carrying out five stages of continuous variation modes, which are respectively the Gaussian variation, Cauchy variation, Pg each dimension opposition-based variation, Pg overall opposition-based variation and DE-based variation; and optimizing the hidden layer unit parameters of the SLFNs by using the optimization strategy of the particle swarm optimization (PSO) algorithm. According to the ELPSO-ELM hybrid intelligent algorithm prediction model disclosed by the invention, the limitations of traditional models which contain less prediction factors and have high data requirements can be solved.

The invention relates to a video content description method based on frame selection, comprising the following steps: a feed-forward neural network is used for constructing a screening model, and thescreening model screens the video frame according to the visual richness and semantic consistency of the video frame; constructing a description model for describing the content of the video to be described; training the screening model and the description model with training data; Selecting a description frame in the video to be described through the filtering model; the visual features of the description frame are extracted and the description model is input to obtain the description sentence of the video to be described.

The invention provides a Chinese phonetic symbol keyword retrieving method based on a feed forward neural network language model. The method comprises: (1), an input sample including historical words and target words are inputted into a feed forward neural network model; for each target word wi, a plurality of noise words with probability distribution q (wi) are added and an active output of a last hidden layer is transmitted to the target words and nodes where the noise words are located, and conversion matrixes between all layers are calculated based on an objective function; errors between an output of an output layer and the target words are calculated, all conversion matrixes are updated until the feed forward neural network model training is completed; (2), a target word probability of inputting a word history is calculated by using the feed forward neural network model; and (3), the target word probability is applied to a decoder and voice decoding is carried out by using the decoder to obtain word graphs of multiple candidate identification results, the word graphs are converted into a confusion network and an inverted index is generated; and a keyword is retrieved in the inverted index and a targeted key word and occurrence time are returned.

The present invention discloses a gesture identification system based on a Chebyshev feed forward neural network. The system comprises a signal emission module, a signal receiving module and a signal pre-processing module which are connected in order, and an attribute characteristic vector extraction module connected with the signal pre-processing module. The attribute characteristic vector extraction module is connected with a Chebyshev feed forward neural network classifier; the signal emission module is configured to emit ultrasonic signals; the signal receiving module is configured to receive the reflected ultrasonic echo signals; the signal pre-processing module is configured to perform pre-processing of the received ultrasonic echo signals; and the attribute characteristic vector extraction module is configured to extract the attribute characteristic vectors of the gesture motions; and the Chebyshev feed forward neural network classifier is configured to identify the attribute characteristic vectors and output identification results. The gesture identification system and method based on a Chebyshev feed forward neural network are able to perform accurate identification of different users' gestures at different environments.

The invention discloses a service life prediction method for an electromechanical system and critical components under the completely truncated data condition. The service life prediction method comprises firstly, extracting originally completely truncated data features based on wavelet packet decomposition to obtain a feature vector sequence set of the completely truncated data; then, calculating a truncation minimum quantitative error (MQE) sequence based on feature fusion of a self-organizing feature map (SOM); performing chaotic parallel multi-layer perceptron (CPMLP) modeling and determining an extending MQE sequence and the failure time based on the CPMLP model; and finally, constructing a feed forward neural network (FFNN) target vector based on survival probability calculation of an intelligent product-limit estimator (iPLE) and performing FFNN training and testing and service life prediction. According to the method, performance degradation indicators of the electromechanical system and critical components are established, service life estimation values of predicted objects are obtained by using fitting residuals, survival probabilities of the predicted objects in a period of time in the future are obtained, and the completely truncated data problem facing the service life prediction of the electromechanical system and critical components is solved.

The invention discloses a big-data-mining-based tool wear state prediction method under a variable working condition. The method is used for solving the technical problem that an existing tool wear state prediction method is poor in accuracy. By means of the technical scheme, the large data technology is adopted, complete sample data under the variable working condition factor are obtained, the FFNN data mining method is improved, the incremental learning capacity is achieved, the new condition can be continuously fused, and the more accurate predication model is obtained. Due to consideration of the complete sample data affecting tool wear and incremental learning of the new tool state feature vector, the new condition is continuously fused and learnt, the more accurate prediction model is obtained, and the basis is provided for further analysis of relevant factors affecting tool wear. Compared with the prior art, the newly-added condition is learnt, and accuracy of the tool wear state predication under the variable working condition is improved. Through detection, according to the minimum error result, the err is equal to 0.03933, and the predication accuracy rate is improved by more than 96%.

The invention discloses a method for automatically classifying intelligent contracts on the basis of feed-forward neural networks. The method has the advantages that intelligent contract codes can be automatically classified by machines; the time and economic costs can be saved as compared with current traditional methods for manually classifying existing intelligent contracts, and the method is competent for tasks with massive contracts which cannot be manually classified; re-training can be carried out by the aid of updated contract sets owing to training modes of the feed-forward neural networks when labeled intelligent contract sets are updated, so that automatic classifiers can adapt to new change.

The invention discloses a three-dimensional myocardial deformation strain calculation method. The invention divides the myocardium into a plurality of local (sub) regions according to the time and space coordinates on the basis of completing the segmentation processing of the marked cardiac magnetic resonance image sequence, a feed forward neural network (BPNN) and a polynomial or support vector machine (SVM) are used for fitting a local displacement field, the calculation of the established local continuous displacement field is done by the Newton iterative method, the motion parameters of the myocardial particle are calculated and the non-linear interpolation technology is used for calculating the myocardial strain parameters. The invention has clear physical significance and simple and effective algorithm, a model is applicable to parallel calculation, which can calculate the forward and backward motion of arbitrary myocardial particle; and the strain calculation results can be used as the judgment criteria of the three-dimensional cardiac ultrasound strain measurement results.

The invention discloses an RCS near-far field transformation method based on a deep neural network, which comprises the steps of first, selecting a neural network according to radar echo data measuredin the near field, wherein a feed-forward neural network is selected if the radar echo data is single-frequency-point data, and a convolutional neural network is selected if the radar echo data is multi-frequency-point data; second, obtaining near-field RCS data and corresponding far-field RCS data to serve as training samples, wherein the near-field RCS data serves as input of the neural network, an output expected result is compared with the generated far-field RCS data, the neural network is trained through an error back-propagation algorithm, and a neural network conforming to error requirements is obtained through adjusting control parameters of the neural network; and third, inputting the RCS data measured in the near field into the trained neural network in actual transformation soas to obtain the transformed far-field RCS data. The RCS near-far field transformation method reduces numerical values caused by the discretization in implementation of the traditional algorithm, andis a brand new RCS near-far field transformation method.

A method of controlling engine NOx production is provided. The method may include determining a desired amount of NOx production for at least one engine cylinder at a first time and determining at least one engine operating parameter to produce the desired amount of NOx using a feed-forward neural network.

The invention provides a material recognition method based on texture surface contact acceleration touch sense information, and belongs to the fields of electronic information, artificial intelligence, mode recognition and machine learning. The method includes the steps of: 1) dividing objects into different material subclasses, selecting objects corresponding to the different material subclasses, and collecting triaxial acceleration data to configure a training sample data set; 2) extracting features from the training samples to obtain fusion feature vectors of the material subclasses, which form a fusion feature matrix; 3) training a feed-forward neural network by means of the matrix, and collecting the triaxial acceleration data of a to-be-tested object and extracting the features, feeding the features to the feed-forward neural network, and predicting and outputting a subclass material, which is corresponding to a maximum value in the matrix, by the network, thereby obtaining the material analysis result of the to-be-tested object. The device includes a vibration sensing body, a data collection module and an upper computer. In the method, by acquiring the contact acceleration information of the surface texture of a commodity, the material of the commodity is determined. The method and the device are applied to internet shopping and can easily, accurately and effectively reflect the actual situation of a commodity.

The invention provides a short-term wind power prediction method based on a cloud evolutionary particle swarm algorithm, including the following steps: S1, building a feed-forward neural network prediction model; S2, measuring the wind speed and direction at a prediction position; S3, getting an initial prediction solution set of the output power of a wind turbine; S4, forming an initialized particle swarm; S5, building a cloud evolution model; S6, generating a first updated particle swarm through the cloud evolution model; S7, updating the first updated particle swarm to get a second updated particle swarm; S8, judging whether the second updated particle swarm satisfies an expected value; if the second updated particle swarm satisfies the expected value, taking the second updated particle swarm as an output result; and if the second updated particle swarm does not satisfy the expected value, continuing to perform the subsequent step; and S9, judging whether the current number of iterations reaches a preset maximum value of iterations; if the current number of iterations reaches the preset maximum value of iterations, taking the second updated particle swarm as an output result; and if the current number of iterations does not reach the preset maximum value of iterations, returning to S6. The short-term wind power prediction method based on a cloud evolutionary particle swarm algorithm has the advantages of high accuracy, good stability and high efficiency.

The invention belongs to the technical field of radio signal positioning, and particularly relates to a multi-target direct positioning method based on neural network computing. The multi-target direct positioning method includes the steps: firstly, decomposing an interested target area into a plurality of sectors, and detecting an existed target sector by the aid of a multilayer feed-forward neural network; secondly, further decomposing the sector into a plurality of sub sectors when a plurality of targets in some sector is detected to ensure at most one target in each sub sector, and detecting the sub sectors with the targets by the aid of the multilayer feed-forward neural network again; thirdly, sequentially performing spatial filtering on the sectors or sub sectors with the targets bythe aid of the multilayer feed-forward neural network; finally, independently, parallelly and directly positioning the targets in different sectors or sub sectors by the aid of a radial basis neuralnetwork.

A post processing system for electrical impedance tomography (EIT) images includes a processing device and a post processing device, and the post processing device is coupled to the processing device. The processing device is configured to generate a first EIT image through a solving method based on the measuring data. The measuring data is measured by an electrical impedance tomography instrument. The post processing device is configured to receive the first EIT image and post-process the first EIT image through a neural network algorithm to generate a second EIT image. The neural network algorithm is a feed forward neural network, a recurrent neural network, a convolutional neural network or a deep neural network, and an accuracy of the second electrical impedance tomography image is higher than an accuracy of the first electrical impedance tomography image.

The invention discloses a method for identifying the fault degree of a high-pressure heater water supply system and belongs to the technical field of thermal system fault diagnosis. The method comprises steps of obtaining fault samples of different types and different severity degrees and performing standardization processing; establishing a probabilistic neural network fault diagnosis model using fault samples of different types; establishing a feed-forward neural network degree identification model for faults of each type; inputting real-time fault sample data in the probabilistic neural network fault diagnosis model and outputting a fault type; selecting a feed-forward neural network degree identification model corresponding to the fault type; continuously inputting the real-time fault sample data in the selected feed-forward neural network degree identification model, and a feed-forward neural network identifying and outputting the fault severity degree; and displaying the fault type and the fault severity degree. The diagnosis speed is fast and the identification precision is high. The fault severity degree can be effectively identified, and the value of the fault severity degree can be provided. The method can be used for high-pressure heater water supply system fault diagnosis in a rated condition, different steady-state conditions and variable conditions.

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.

Systems, methods, and computer program products are provided to provide noise reduction for an input signal using a neural network. A feed-forward set of neuron groups is provided to enhance neuron activity within a particular frequency band based on prior reception of activity within that frequency band, and also to attenuate surrounding frequency bands. A surround-inhibition set of neuron groups further attenuates activity surrounding the stimulated frequency band.

The invention discloses a design method of an intelligent rotary table control system based on adaptive dynamic planning. The design method comprises the following steps: first, the angular speed deviation value delta theta<L>(t), namely the deviation between the actual rotary table angular speed and the control required angular speed, of a rotary table is determined and selected as the system state quantity, two delay angular speed deviation values delta theta<L>(t-1) and delta theta<L>(t-2) of the rotary table are determined, and the motor end voltage Ua (t) is selected as the control quantity; second, two multi-layer feed-forward neural networks, namely the execution network and the evaluation network are constructed, and each network is only provided with one hidden layer; third, an intelligent controller algorithm is edited; and fourth, a hardware system is connected to form online learning of a closed-loop control system, wherein as for the closed-loop control system, an upper computer intelligent algorithm controller is used for control, a frequency converter is used for driving, and the rotary table system freely rotates and feeds the angular speed quantity back to an uppercomputer. The low-speed performance of the rotary table system can be effectively improved, and the dynamic range of the rotary table system can be effectively increased.

The invention provides a Chinese phonetic symbol keyword retrieving method based on a feed forward neural network language model. The method comprises: (1), an input sample including historical words and target words are inputted into a feed forward neural network model; for each target word wi, a plurality of noise words with probability distribution q (wi) are added and an active output of a last hidden layer is transmitted to the target words and nodes where the noise words are located, and conversion matrixes between all layers are calculated based on an objective function; errors between an output of an output layer and the target words are calculated, all conversion matrixes are updated until the feed forward neural network model training is completed; (2), a target word probability of inputting a word history is calculated by using the feed forward neural network model; and (3), the target word probability is applied to a decoder and voice decoding is carried out by using the decoder to obtain word graphs of multiple candidate identification results, the word graphs are converted into a confusion network and an inverted index is generated; and a keyword is retrieved in the inverted index and a targeted key word and occurrence time are returned.

The invention provides a livestock delivery detecting method. The livestock delivery detecting method comprises the steps of, firstly, collecting data before and after livestock delivery; secondly, through the collected data and delivery results corresponding to the data, training a delivery detecting model, which can automatically select channel data facilitating delivery discrimination and provide a discrimination result through the channel data; thirdly, inputting newly-collected data with the answer unknown, and performing discrimination through the delivery detecting model to acquire a delivery result. According to the livestock delivery detecting method, a long short-term memory neural network model and a first feedforward neural network model cooperate to automatically select channels facilitating detection; a second feedforward neural network model is applied to selecting channel data with high data discrimination capacity, which is applied as the final basis for determining the delivery result, so that the delivery detecting precision can be improved. The livestock delivery detecting method is wide in application range and applicable to various complex situations.

A post processing system for electrical impedance tomography (EIT) images includes a processing device and a post processing device, and the post processing device is coupled to the processing device. The processing device is configured to generate a first EIT image through a solving method based on the measuring data. The measuring data is measured by an electrical impedance tomography instrument. The post processing device is configured to receive the first EIT image and post-process the first EIT image through a neural network algorithm to generate a second EIT image. The neural network algorithm is a feed forward neural network, a recurrent neural network, a convolutional neural network or a deep neural network, and an accuracy of the second electrical impedance tomography image is higher than an accuracy of the first electrical impedance tomography image.

The invention provides a cross-media analysis and reasoning fire detection algorithm, and the algorithm comprises fire scene identification, feature extraction, region segmentation, feature fitting, movement estimation and a plurality of acquisition devices; the feature fitting is used for identifying a movement path of a fire; the movement pre-estimation is based on the data of the feature fitting, and the movement path of the fire area is pre-estimated; each acquisition device acquires fire scene data for performing position identification on a fire scene; the area of the fire scene is recognized based on a feedforward neural recognition network, feature extraction is performed on the recognized area of the fire scene, and the range of fire movement is estimated. According to the method, the fire scene is identified by adopting the search ring, and nonlinear fitting is carried out by combining with the moving path of the fire behavior, so that the fire passing range of the fire scene can be accurately identified; meanwhile, analysis is carried out in combination with the eccentricity and the circularity, so that the ignition analysis of the fire scene is more accurate.

The invention discloses a method and a system for detecting voice spoofing attack of speakers on the basis of deep learning. The method includes constructing audio-frequency training sets, initializing deep feed-forward neural networks and deep recurrent neural networks and respectively training the deep feed-forward neural networks and the deep recurrent neural networks by the aid of multi-frame feature vectors and single-frame vector sequences of the training sets; respectively leading frame level and sequence level feature vectors of to-be-tested audio frequencies into two trained linear differential analysis models in test phases, weighting two obtained result grades to obtain scores and comparing the scores to predefined threshold values so as to discriminate voice spoofing. The method and the system have the advantages that local features can be captured, and global information can be grasped; the linear differential analysis models are used as classifiers in identification and verification phases, the voice spoofing attack can be judged by means of grade fusion, and accordingly the voice spoofing detection accuracy can be greatly improved.