3589 results about "Neutral network" patented technology

The present invention relates to an intelligent adaptive system and method for monitoring leakage of oil pipeline networks based on big data. The present invention effectively analyzes a large amount of data collected on site within a reasonable time period and obtains a state of a pipeline network by an intelligent adaptive method, thereby obtaining a topological structure of a pipeline network. The present invention specifically adopts a flow balance method in combination with information conformance theory to analyze whether the pipeline network has leakage; small amount of leakage and slow leakage can be perfectly and accurately alarmed upon detection; as a generalized regression neural network is adopted to locate a leakage of the pipeline network, an accuracy of a result is increased. Therefore, the present invention adopts a policy and intelligent adaptive method based on big data to solve problems of detecting and locating leakage of the pipeline network.

A method and system for early detection of incipient faults in an electric motor are disclosed. First, current and voltage values for one or more phases of the electric motor are measured during motor operations. A set of current predictions is then determined via a neural network-based current predictor based on the measured voltage values and an estimate of motor speed values of the electric motor. Next, a set of residuals is generated by combining the set of current predictions with the measured current values. A set of fault indicators is subsequently computed from the set of residuals and the measured current values. Finally, a determination is made as to whether or not there is an incipient electrical, mechanical, and / or electromechanical fault occurring based on the comparison result of the set of fault indicators and a set of predetermined baseline values.

The invention discloses an auxiliary film reading system for a capsule endoscope image. A data acquisition module is used for acquiring capsule endoscopic image data of an examined patient; an image position classification module is used for classifying capsule endoscopic images according to different shot parts by using a first convolution neural network (CNN) to obtain image sequences of different shot parts; an image sequence description module is used for carrying out image feature extraction on the image sequences of different shot parts by using a second convolution neural network to obtain feature vector sequences of different alimentary canal part image sequences and for transforming image features in the feature vector sequences into descriptive texts by using a recursive neural network (RNN) to generating an auxiliary diagnosis report. Therefore, the work load of observing an alimentary canal image by a doctor can be reduced and thus the diagnosis efficiency of the doctor can be improved.

A migration learning lung lesion tissue detection system based on an MaskScoring R-CNN network comprises a storage module for storing four lung diseases including lung cancer, pneumonia, pulmonary tuberculosis and emphysema and further comprises a diagnosis module, and the diagnosis module is in communication connection with the storage module and comprises the following steps of 1) preprocessinga medical image; 2) constructing the MaskScoring R-CNN network model, wherein the step 2) specially comprises 1, constructing a shared convolutional neural network backbone (for feature extraction); 2, carrying out transfer learning on a shared convolutional neural network; 3, constructing an FPN network; 4, constructing an RPN network; 5, constructing an ROIAlign layer; 6, adding the MaskIoU head; and 3) identifying the lung medical image lesion tissue, inputting a to-be-detected lung CT image into the constructed MaskScoring R-CNN network, outputting and obtaining an identified image by thenetwork, framing out and masking the identified lesion tissues, and marking the lesion categories. According to the method, the requirement for high precision of medical image segmentation is met, andthe network can have the good generalization.

The invention discloses a non-restricted environment face verification method based on block depth neural network. The method comprises the following steps of (1) detecting a face area at which a face image is input, and normalizing the face area; (2) dividing the normalized face area into a plurality of non-overlapping rectangular subimages, extracting feature of each subimage, and performing dimensionality reduction and normalization processing; (3) building one depth neural network for each subimage according to the extracted subimage features, wherein the subimage features are changed into new features after being input into network; (4) according to paired face image data and the depth neural network group, optimizing structure parameter of the depth neural network by restraining foreign separability and congeneric compactness of the changed new features; and (5) inputting paired face images into the optimized depth neural network group, calculating distance between the new features, and verifying the face pair.

An automated kitchen assistant system inspects a food preparation area in the kitchen environment using a novel sensor combination. The combination of sensors includes an Infrared (IR) camera that generates IR image data and at least one secondary sensor that generates secondary image data. The IR image data and secondary image data are processed to obtain combined image data. A trained convolutional neural network is employed to automatically compute an output based on the combined image data. The output includes information about the identity and the location of the food item. The output may further be utilized to command a robotic arm, kitchen worker, or otherwise assist in food preparation. Related methods are also described.

The invention discloses a text multi-label classification method based on semantic unit information, which comprises the following steps: establishing a semantic unit multi-label classification modelSU4MLC, taking a recurrent neural network sequence based on an attention mechanism to a sequence model as a baseline model for improvement, and improving the expression of the attention mechanism by improving a source end; Extracting semantic unit related information from the context representation of the source end of the baseline model by using hole convolution in deep learning to obtain semantic unit information; Combining the semantic unit information with the word level information by using a multi-layer mixed attention mechanism, and providing the combined information for a decoder; Anddecoding the tag sequence by using a decoder, thereby realizing text multi-tag classification based on semantic unit information. According to the method, the problems that an existing attention mechanism is easily influenced by noise and contributes to classification insufficiently can be solved, the contribution of the attention mechanism to text classification can be improved, and the text multi-label classification problem can be more efficiently solved.

Measurements of fluorescence spectra of fluid samples recovered downhole are processed to give the fluid composition. The processing may include a principal component analysis followed by a clustering method or a neutral network. Alternatively the processing may include a partial least squares regression. The latter can give the analysis of a mixture of three or more fluids.

The invention provides a face attribute recognition method based on a combined depth neural network. The method comprises the following steps: S1, performing area extraction on an inputted face image;S2, carrying out positioning on a facial important part and extracting a related area based on the area extraction result to obtain facial area features and facial key part features; S3, marking a real age and a gender of each face by using the area extraction result as a training sample, carrying out training based on a deep neural network, and outputting an age estimation value and a gender recognition result; and S4, on the basis of the facial area features outputted by the S2 and local area features of eyes and mouths, outputting expression and wearing prediction probabilities respectively according to a random forest algorithm, adding probabilities of corresponding attributes for different attributes, and outputting facial expression and wearing attribute results. In addition, the invention also provides a face attribute recognition apparatus based on a combined depth neural network.

The invention provides a power battery SOC estimation method, wherein power battery SOC estimation is performed based on a BP neural network assisted extended Kalman filter. According to the power battery SOC estimation method, a state estimation updated value is used as the input value of the BP neural network, and the estimated value of an observation noise variance-covariance matrix is used as an objective output value of the BP neural network, thereby performing online training on a constructed BP neural network. The observation noise variance-covariance matrix which is output by the BP neural network is supplied to an error variance-covariance prediction equation and a filtering gain equation of the extended Kalman filter, thereby realizing recursive calculation of the BP neural network assisted extended Kalman filter. The power battery SOC estimation method can settle the problems such as incapability of satisfying a requirement for online estimation, large accumulative error, easy diffusion, and easy influence by a noise in an existing estimation method. Furthermore the power battery SOC estimation method has high estimation precision.

The invention discloses a near infrared spectrum analyzing method based on the isolated component analysis and genetic neural network, which comprises the following steps for the acquired near infrared spectrum: firstly, effectively compressing spectrum data by using wavelet transform; secondly, extracting an independent component and a corresponding mixed coefficient matrix of a near infrared spectrum data matrix by using an isolated component analysis method; thirdly, building a three-layer BP neutral network, using the mixed coefficient matrix of a training sample as the input and correspondingly measured component concentration matrix as the output, and optimizing a neutral network structure by adopting a genetic algorithm, and obtaining a GA-BP neutral network by the training of the training sample; fourthly, predicting and analyzing the measured component concentration of the predicted set sample by using the GA-BA neutral network. The method enriches the chemical measurement method, widens the application range of the isolated component analysis and has favorable application prospect.

The invention discloses an automatic question generation method based on deep learning. The method comprises the following steps: constructing a training set (articles, answers and questions), a verification set (articles, answers and questions) and a prediction set (articles and answers); creating Encoder-based construction by using deep learning framework tensorflow A sequence of decoders to a sequence neural network model; Carrying out word segmentation, word list making and word embedding operations on sentences in the data set; Wherein the data set comprises a training set, a verificationset and a prediction set; The training set is used for training the model, the verification set is used for detecting whether the currently trained model is over-fitted or not, and if over-fitting isachieved, stopping training; Otherwise, continuing training; And decoding the prediction set by using the trained model to generate a problem. The method is good in generalization effect and low in labor cost, the generated questions are better matched with articles and answers, and the method can be widely applied to the fields of intelligent teaching, intelligent questions and answers, knowledge question and answer games and the like.

An implantable apparatus, such as a cochlear implant, for delivering electrical plasticity informative stimuli to a neural network of an implantee. The apparatus comprises a stimulator device (40) that generates stimulation signals, and an electrode array (20) that receives the stimulation signals and delivers the stimuli to the neural network of the implantee in response to the signals. The stimuli delivered to the implantee facilitates and / or controls the production and / or release of naturally occurring agents into the neural network to influence the functionality thereof.

The invention discloses a debris flow risk degree evaluation method. The method comprises a step of determining the characteristic parameters associated with debris flow three elements, a step of establishing the comprehensive information evaluation system of the three elements, obtaining a three-element initial information evaluation matrix, and calculating the information entropy of the three elements through matrix operation and an entropy method, a step of establishing a debris flow information entropy model, taking three sub information entropy as an input factor, through BP neural network effect, and outputting debris flow information entropy, a step of defining a debris flow risk degree level standard according to the relation between the information entropy theory and the fact whether debris flow occurs, and carrying out risk degree evaluation on a search object. The invention discloses a debris flow probability model, the mutual interaction mechanism of debris flow three elements in a debris flow inoculation process can be comprehensively reflected, the complex nonlinear and dynamic processes of debris flow can be represented, the risk degree of the research object (single channel / regional debris flow) can be forecasted.

The invention discloses a wearable device based on neural network adaptive health monitoring. The wearable device comprises an exercise state solving module, an exercise link judgment module, a dimension reduction solving module, a physiological parameter setting module and a physiological parameter collection module; the exercise state solving module collects acceleration data output by an acceleration sensor, high-frequency noise is filtered through wavelet transformation, and the exercise state of a user is divided; the exercise link judgment module subdivides the exercise state in combination with an angular velocity to obtain an exercise subdivision link; and according to the dimension reduction solving module, an optimized SVM is used to perform further fine division on partial pathological features hierarchically in combination with data at different levels. According to the wearable device, different feature dimension reduction samples are integrated in an analysis link, the amount of exercise is greatly reduced, correctness is guaranteed by emphatically calculating a verification link, the medical reference value of obtained user physiological parameters is increased, and power consumption standby capability is improved under the same condition.

The invention discloses an image compression system, a decompression system, a training method and device, and a display device. In the image compression system, a convolution neutral network module is utilized to complete updating and predicting processes, so that a corresponding image compression system has a better compression ratio when weight of each filter unit in the convolution neutral network module is trained. Therefore, the difficulty of setting filtering parameters of an image compression unit and an image decompression unit is reduced.

The invention discloses a behavior intention fused surrounding dynamic vehicle trajectory prediction system and method. The system comprises a trajectory prediction module and a behavior intention prediction module, the trajectory prediction module takes information of historical trajectories of a target vehicle needing trajectory prediction and vehicles around the target vehicle as input of a long-short-term memory regression neural network, and prediction trajectories of a future time domain are obtained through network prediction; the behavior intention prediction module is used for obtaining probability distribution of behavior intentions obtained based on prediction tracks of a target vehicle and surrounding vehicles by considering behavior interaction between different vehicles and utilizing an LSTM classification neural network; the results of the two modules are fused and input into a multi-modal LSTM trajectory prediction neural network to obtain the position information of the final prediction trajectory. According to the method, the motion information of the vehicle and the information of the surrounding traffic environment are fully utilized, the dynamic change and uncertainty of the traffic environment are considered, the accuracy of trajectory prediction is improved, and the system and method are suitable for more complex driving scenes.

A method for monitoring a person performing a physical exercise based on a sequence of image frames showing the person's exercise activity is described. The method comprises the steps of extracting, based on the sequence of image frames, for each image frame a set of body key points using a neural network, the set of body key points being indicative of the person's posture in the image frame, and deriving, based on a subset of the body key points in each image frame, at least one characteristic parameter indicating the progression of the person's movement. The method further comprises detecting a start loop condition by evaluating the time progression of at least one of the characteristic parameters, said start loop condition indicating a transition from a start posture of the person to the person's movement when performing the physical exercise.

The invention discloses a prisoner psychological health state assessment method and a system based on multi-modal information. The method comprises the following steps: obtaining physiological signals, facial expression images and voice signals of reformed prisoners and prisoners to be tested after they experience virtual reality scenes, and extracting physiological signal features, facial expression image features and voice signal features from the obtained signals; inputting the physiological signal features, facial expression image features and voice signal features of the reformed prisonerinto a pre-trained neural network model, and outputting a psychological state assessment vector of the reformed prisoners; inputting the physiological signal features, the facial expression featuresand the voice signal features of the prisoners to be tested into a pre-trained neural network model, and outputting a psychological state assessment vector of the prisoners to be tested; calculating the psychological state assessment vector distance between the prisoners to be tested and the reformed prisoners; and assessing the mental health state of the prisoners according to the distance.

The invention relates to a preceding vehicle following method based on a deep convolutional neutral network and solves the problems that a self-adaptive cruise system in the prior art has tedious visual ranging and controller design processes, cannot fine adjust control parameters according to driving habits of drivers and cannot meet different personal driving habits. Driving behavior data are trained, and under the condition of given visual sensing input, accurate control of vehicle accelerators and automatic pedals under multiple work conditions is realized by imitating human drivers according to feature description of the surroundings by the deep convolutional neutral network. All that is required is to fine adjust the deep neutral network parameters with different work condition training samples instead of designing the controller structure and adjusting the controller gain under different work conditions. The tedious visual ranging and controller design processes are avoided, and the vehicle following behavior of the drivers is simulated by adjusting the deep neutral network parameters.

The invention relates to a tide predicting method for the tide is influenced by various factors, including cyclical factors, such as tidal generation force, and non-cyclical factors, such as wind power, atmospheric pressure, coast characteristics, rainfall, dip angles of the lunar orbit and the like. The predicting accuracy of the traditional harmonic analysis method is influenced by partial tide number, and the traditional harmonic analysis method cannot analyze the influence of non-cyclical factors; the artificial neural network method developed recent years overcomes the defect that the non-cyclical factors cannot be predicted by the harmonic analysis method to a certain extent, but has great data volume required by study training samples and wide involve range, can cover various possible conditions, but has less station historical data of non-cyclical factors. The invention provides a predict model, wherein factors which influence tide non-cyclically, such as wind directions, rainfall, storm surge, coast characteristics and the like, can be fused into the model, and small sample data can receive more accurate results. In the method, a support vector machine (SVM)-based predict model is established, wherein, an SVM toolbox is imported into MATLAB 7.8; training sample data is trained by utilizing svmtrain function; the formed model is tested by using a test sample svmpredict function; and the trained and tested data can predict the tide in the same tide test station.

A triaxial magnetic electronic compass error compensation method based on depth learning. An implicit error model is trained for compensating non-linear error in magnetic compass measurement and improving the orientation accuracy of the magnetic compass. The error model training consists of two stages: a first stage is pre-training, and a second stage is reverse trimming by using a back-propagation algorithm for fine-tuning all layers of the network, and reducing the error of model training. The magnetic compass calibration and compensation procedure is to use depth learning algorithm training to obtain a non-linear error model, and the distorted measurement magnetic field is inversed back to a true magnetic field value, thereby reducing the calculation error of course angle. The invention aims at nonlinear error of magnetic compass and provides the error training method based on depth learning; compared to random initialization of a traditional neural network, the weight of each layer locates in a better position of parameter space, so as help to improve the convergence of the algorithm and model training accuracy and achieve high-precision orientation of magnetic compass.

A tunnel fire early-warning controlling method based on multi-sensor data fusion technology and a system using the tunnel fire early-warning controlling method based on the multi-sensor data fusion technology comprise steps as below: recording of history data of a main sensor group, reading of data of an auxiliary sensor group at a regular time, calculation of a neural network, passing back of a calculation result, execution controlling on site and the like, and the system specially using the tunnel fire early-warning controlling method based on the multi-sensor data fusion technology comprises a site detection device, a data processing device and an alarming displaying device. The tunnel fire early-warning controlling method based on the multi-sensor data fusion technology comprises a dynamic and flexible fire disaster threshold value adjusting method, namely, a method calculating with a neural network to analyze current data of the history data of the main sensor group and the auxiliary sensor group to obtain a fire disaster threshold value of current environment factors, and the original fire disaster threshold value is replaced by the new fire disaster threshold value. The tunnel fire early-warning controlling method based on the multi-sensor data fusion technology can effectively reduce judgment differences of a fire disaster caused by changing of environment of the tunnel, and reduce misstatement rate and missing reporting rate of the funnel fire disaster early-warning system.

The invention discloses an athlete athletic injury risk early warning method. On the bases of reference to models provided by foreign scholars and comprehensive analysis of track and field athlete athletic injury factors, a track and field athlete motion injury risk early warning injury factor dynamic chain model is provided, based on the model, corresponding factors are selected from an athlete risk early warning database, and a track and field athlete motion injury risk early warning injury factor dynamic chain quantitative model is established through the analytic hierarchy process. Index data are discretized according to an SOM neutral network discretization method, a decision table is simplified according to a method based on a discernibility matrix in a rough set, an RBF neutral network is established based on the simplified decision table, the RBF neutral network is trained, and finally a correct diagnosis result is acquired. By the adoption of the method, accurate early warning can be conducted on athletic injuries, the athlete athletic injury risk grade is effectively predicted, and treatment and prevention of the athletic injuries are facilitated.

This invention relates to deficiency identification and quantification based on three dimensional limit element neutral network, which comprises the following steps: a, according to deficiency leakage magnetic field three dimensional element computation module forming three dimensional element neutral network; b, measuring and extracting deficiency magnetic field characteristics value and setting measurement values and computing error valve conditions; c, given deficiency characteristics parameters initial values by use of three dimensional limit element neutral network for overlap computation to realize deficiency identification and evaluation through comparing deficiency computation values and error size.

The invention discloses a BP neural network construction and identification method for identifying different brown rice grains, adopting the following technical scheme: including the following steps: 1) acquiring images: 2) image preprocessing: 3) extracting image feature information of different brown rice grains: 4) main Component analysis reduces dimensionality of image feature information; 5) Design BP neural network structure: 6) Train neural network, and use the BP network neural network constructed by any of the above-mentioned construction methods to identify different brown rice grains. The method proposed by the present invention obtains images of different types of brown rice grains through black as the background, uses image processing technology to obtain its characteristic information, and uses principal component analysis to reduce the dimensionality of the characteristic information, and finally uses BP neural network to perform different types of brown rice grains. identify. The method can identify different types of grains objectively, accurately and quickly, and overcomes the disadvantages of traditional manual detection.

The present invention provides a simple, robust, and versatile high-resolution spectrometer that is suitable for downhole use. The present invention provides a method and apparatus incorporating a spinning, oscillating or stepping optical interference filter to change the angle at which light passes through the filters after passing through a sample under analysis downhole. As each filter is tilted, the color or wavelength of light passed by the filter changes. Black plates are placed between the filters to isolate each filter's photodiode. The spectrometer of the present invention is suitable for use with a wire line formation tester, such as the Baker Atlas Reservation Characterization Instrument to provide supplemental analysis and monitoring of sample clean up. The present invention is also suitable for deployment in a monitoring while drilling environment. The present invention provides a high resolution spectometer which enables quantification of a crude oil's percentage of aromatics, olefins, and saturates to estimate a sample's gas oil ratio (GOR). Gases such as CO2 are also detectable. The percentage of oil-based mud filtrate contamination in a crude oil sample can be estimated with the present invention by using a suitable training set and chemometrics, a neural network, or other type of correlation method.

The invention discloses a flood forecasting method based on cluster analysis and real time correction, which comprises the following steps: 1) using PCA(Principal Component Analysis) to perform dimensionality reduction to the input of a model; 2) using the K-means clustering method to conduct clustering analysis on original data; dividing the flood data into different classifications; and then training different SVM models; when a testing sample is inputted, using the clustering center to determine the classification of the test sample and predicting the corresponding model to obtain a predicted value q; and 3) using a BP neural network for real time correction; calculating the error sequence between the predicated value and the actual value; using the error sequence data to train the BP neural network error correction model to obtain the error correction value qe. The final forecasting result is the model predicted value q plus the error correction value qe. According to the invention, the original hydrological data are divided into several classifications by cluster analysis, and through the training of the models, forecasting can be available by the multiple models. Then, real-time correction is achieved by the BP neural network to improve the forecasting accuracy for the time of flood peak.

The invention discloses a dispute focus automatic identification method based on a hierarchical attention neural network model. The method comprises the steps of firstly, extracting a dispute focus statement of a court for case induction from a document containing a dispute focus induced by the court, and constructing a dispute focus system by utilizing a hierarchical clustering method; and marking a plurality of different category labels for each document by utilizing the dispute focus system, constructing a data set, and converting a dispute focus identification problem into a multi-label multi-classification problem; then, training a hierarchical attention neural network model, paying more attention to important words, sentences and paragraphs containing more information, and forming adispute focus recognizer; and finally, inputting the text of which the dispute focus needs to be identified into a dispute focus identifier to obtain the dispute focus of the input text. The method ishigh in prediction accuracy, can accurately identify and judge the dispute focus of the document, and has good expandability.

The invention discloses an event type recognition method and device. The method comprises the following steps that word segmentation is performed on all texts in a training set, a word vector space model is trained after word class treatment is extracted, text characteristics are extracted, and texts are shown as characteristic vectors; even type clustering is performed on the training set, and a neutral network model with type cluster regularization items is trained; test samples are also analyzed, word class treatment is extracted, and the trained word vector model is trained for obtaining character representations; by means of the neutral network model with the type cluster normalization items, event type recognition is performed. By means of the technical scheme, problems brought by annotation data imbalance are relieved by means of type sharing information in the same group.