44results about How to "Effective feature extraction" patented technology

The invention relates to a stable vision-induced brain-computer interface-based robot control method. The method comprises the steps of firstly preprocessing an electroencephalogram signal obtained by actual collection with bandpass filtering; secondly, performing fast independent component analysis on the electroencephalogram signal obtained by preprocessing to obtain independent components; then using a Hilbert-Huang transform to dissolve the independent components to obtain an intrinsic mode function; performing spectrum analysis on the intrinsic mode function to obtain the needed features; finally using a threshold value judging method to classify the extracted features, and translating the classifying result into signals capable of being identified by a robot, thus realizing the real-time control on the robot. The robot control method is based on a stable vision-induced brain-computer interface, is high in transmission rate and is simple in equipment and device. The independent component analysis and Hilbert-Huang transform are combined in the feature extracting process, so the feature extracting is more effective. The limb-free action control on the motion of the robot is realized, and the severely paralyzed disabled with normal brain functions can control the robot to assist the disabled in normal living.

The invention discloses an objective positioning and classifying algorithm based on deep learning. The algorithm comprises the following steps: S1, inputting a picture into a first network and outputting a series of objective positioning frames and scores; S2, inputting a picture and a series of sub windows into a second network; S3, forwardly propagating the networks to a last convolution layer to generate a characteristic pattern; S4, performing coordinate transformation on the sub windows by using a zooming coefficient in order to map the coordinates to the characteristic pattern; S5, acquiring a characteristic from the characteristic pattern by using the zoomed sub windows and pooling the characteristic to a regular size; S6, classifying the pooled data to obtain a classification result and a score of an area; and S7, inputting the objective positioning frames and the classification result of the area into a classifier to carry out classification, and outputting the type and the coordinates of the objective. The algorithm may extract characteristics, classifies the objectives, performs positioning and identification by using a unified network, and greatly increases calculating speed.

The invention discloses a target recognition and shape retrieval method based on hierarchical description. The method comprises the following steps of: extracting the profile feature of a target by a profile extracting algorithm, calculating a curvature value of each point on the profile target, extracting the angular point feature of the target by non-maximum value suppression, taking a profile segment corresponding to every two angular points as an overall feature describer of the target, carrying out hierarchical description on the profile points according to curvature, carrying out hierarchical description on the profile segments according to the importance degrees of value features, combining profile segments, the values of which are lower than evaluation thresholds, to form profile feature segments as partial feature describers of the target, carrying out normalization on the profile feature segments, and carrying out similarity measurement on the profile feature segments of different targets according to Shape Contexts distance. The method can be used for performing feature extraction on a target shape effectively, scale invariance, rotation invariance and translation invariance are achieved, the accuracy rate and the robustness in recognition are improved, and the computation complexity is reduced.

The invention relates to a character-level nested deep network-based text classification method. The method comprises the following steps of S1, constructing a character vector matrix table; S2, performing short text preprocessing; S3, extracting high-dimensional sequence features by improved Resnet; and S4, performing LSTM network classification. The character level-based text conversion can effectively perform conversion on all texts; compared with a conventional vector space model, the dimension is reduced remarkably; all the texts can be effectively converted; and low-frequency words are not ignored. In addition, the improved Resnet can self-learn a feature extraction method; and compared with conventional methods such as a TF-IDF formula, a mutual information quantity, an informationgain, an x2 statistical quantity and the like, the extracted features are more effective and abstract. Finally, the LSTM network classification can consider a sequence relationship between the words,so that the text classification can be performed more accurately.

The invention discloses a multi-modal fusion saliency detection method based on a convolution attention module. The method comprises: in a training stage, constructing a convolutional neural network;inputting the left viewpoint image and the depth image of the original image into a convolutional neural network for training to obtain a corresponding saliency detection image; calculating a loss function between a set formed by saliency detection images generated by the model and a set formed by corresponding real human eye gaze images to obtain an optimal weight vector and an offset term of theconvolutional neural network classification training model; and in a test stage, inputting the three-dimensional image in the selected data set into the trained convolutional neural network model toobtain a saliency detection image. According to the visual saliency detection method, extraction of image features is optimized by applying a novel module, multi-scale and multi-mode feature fusion iscarried out, and finally the detection efficiency and the detection accuracy of visual saliency detection are improved.

The invention discloses an upper ocean thermal structure prediction method based on a deep belief network. The prediction method comprises the following steps of 1) determining proper environment parameters as input factors and predication values of a thermal structure prediction model; 2) establishing an upper ocean thermal structure sample data set, dividing sample data into training data and test data, and performing unified data preprocessing; 3) establishing the deep belief network, and performing non-monitoring pre-training on the sample data layer by layer to obtain relatively excellent parameters of the model preliminarily; 4) based on a back propagation algorithm, performing fine tuning on the parameters of the model according to marks of the training sample to determine optimal parameters; and 5) applying an established upper ocean thermal structure prediction model to test data, and obtaining a predicated temperature value at a specific depth of the upper ocean from the output layer. By adoption of the upper ocean thermal structure prediction method based on the deep belief network provided by the invention, the problems of single fitting and over-fitting of the conventional method are overcome; and a characteristic relation between the ocean surface layer environment parameters and the upper ocean thermal structure is effectively extracted, so that the upper ocean thermal structure prediction accuracy is improved.

A wearing fault diagnosis method of an automobile engine bushing comprises that vibration signals in normal and bushing wearing conditions of an automobile engine are collected, and time-domain statistics feature extraction is carried out; frequency-domain feature extraction is carried out on the vibration signals via a signal processing method of wavelet packet transformation; a kurtosis index ismixed with a wavelet packet energy entropy to obtain a characteristic vector; and a classification model on the basis of a K-nearest neighbor algorithm s established, the characteristic vector is used to train and test the classification model, and thus, the wearing fault of the single bushing of the automobile engine is diagnosed. The method has the advantages that the bushing wearing fault canbe analyzed in focus among common faults of the automobile engine, sufficient and effective characteristic value extraction can be realized for typical faults of the automobile engine, and the fault type can be identified in high accuracy.

The invention provides a training method and device for a brain tumor recognition model based on semi-supervised learning. The brain tumor recognition model comprises a detection network and a classification network, and the method comprises the following steps of obtaining a first training sample set and a second training sample set; performing unsupervised learning on the detection network and the classification network by using the first training sample set to generate a pre-trained detection network and a pre-trained classification network; training the pre-trained detection network and the pre-trained classification network by using the second training sample set to generate a trained detection network and a trained classification network; and outputting a trained brain tumor recognition model, wherein the trained brain tumor recognition model comprises the trained detection network and the trained classification network. According to the method, fine marking and coarse marking data are fully utilized in a semi-supervised learning mode, so a more robust deep convolutional neural network is obtained.

The invention relates to a hyperspectral image classification method, and concretely relates to a hyperspectral image classification method based on spectral space attention fusion and a deformable convolutional residual network. The objective of the invention is to solve the problem of low classification accuracy of hyperspectral images caused by insufficient spectrum and spatial feature extraction and overfitting under small samples due to the fact that the hyperspectral images contain abundant information in the existing hyperspectral image classification. The method comprises the following steps: 1, acquiring a hyperspectral image data set and a corresponding label vector data set; 2, establishing an SSAF-DCR network based on spectrum space attention fusion and a deformable convolution residual error; 3, inputting the x1, the x2, the Y1 and the Y2 into a network SSAF-DCR, and performing iterative optimization by adopting an Adam algorithm to obtain an optimal network; and 4, inputting x3 into the optimal network to carry out classification result prediction. The method is applied to the field of hyperspectral image classification.

The invention belongs to the technical field of aero-engines, and particularly relates to an aero-engine service life prediction method based on improved LSTM. Comprising the steps of 1, processing original sensor data acquired by a sensor, and constructing a training sample which comprises a training set and a test set; step 2, on the basis of constructing the training sample in the step 1, constructing an LSTM structure model as an engine residual life prediction model; and 3, inputting the test set in the step 1 into the LSTM structure model constructed in the step 2 to obtain a predicted RUL value, and evaluating the obtained predicted RUL value by adopting RMSE and Score evaluation indexes. According to the aero-engine life prediction method based on SDAE and LSTM, the advantage of unsupervised feature extraction of a deep encoder is utilized, effective feature extraction is performed on an engine sensor signal, low efficiency of manual feature extraction and prediction uncertainty caused by the low efficiency are avoided, and the prediction accuracy is improved. And the residual life of the engine is predicted by using the advantage of processing the time sequence data by the LSTM model.

The invention relates to a Chinese named entity recognition method, which comprises the steps of obtaining a to-be-recognized statement, inputting the to-be-recognized statement into an embedded layerof a preset neural network, and outputting an output word vector of the to-be-recognized statement; respectively and synchronously inputting the output word vectors into a preset bidirectional long-short-term memory network and a preset hole convolution network to obtain an output matrix; inputting the output matrix into a preset segmented long-short-term memory network to obtain a plurality of prediction sequences; and selecting an optimal sequence from the prediction sequences by adopting a conditional random field algorithm, and obtaining the identified entity according to the optimal sequence. According to the invention, the bidirectional long-short-term memory network and the preset hole convolution network are synchronously adopted for feature transformation, features can be effectively extracted, dependence on an entity dictionary is not needed, and the recognition efficiency is improved.

A method of deriving a representation of an image by processing signals corresponding to the image, the method comprises deriving a function of the image, where the function of a translated, scaled or rotated version of the image is a translated or scaled version of the function of the image, and using a plurality of frequency components of a frequency representation of the function to derive a representation of the image.

The invention discloses a rolling bearing detection method based on LMD (Local Mean Decomposition) and gray correlation, belongs to a rolling bearing fault detection method in the field of mechanical engineering, and relates to a rolling bearing detection method based on a fuzzy entropy algorithm of LMD (Local Mean Decomposition) and gray correlation. The method comprises the steps: employing an acceleration sensor to collect vibration acceleration signals of a rolling bearing during operation, wherein the vibration acceleration signals comprise a no-fault normal bearing vibration acceleration signal and inner ring, rolling body or outer ring fault bearing vibration acceleration signals. carrying out the LMD decomposition of the collected acceleration signals, and obtaining a plurality of PF (product function) components and residual errors; calculating the gray correlation degree of a test sample and a standard matrix through employing a gray correlation algorithm, and carrying out the fault mode recognition. The method can effectively carry out the feature extraction of the vibration signals, solves problems that the EMD decomposition is severe in excessively modal mixing and end effect and a PF component is large in data size and cannot serve as a characteristic vector, and achieves the effective recognition of the operation state of the rolling bearing.

The invention discloses an intelligent brain-controlled mechanical arm auxiliary feeding system and method for disabled people. A human-computer interaction interface module is used for providing a platform for information interaction between a user and a computer system; a signal acquisition module is used for acquiring electroencephalogram data (SSVEP) induced by stimulation stroboflash on a steady-state visual stimulation interface; a signal processing module is used for preprocessing and classifying electroencephalogram signals acquired by the signal acquisition module and converting a classification result into a control instruction; a mechanical arm feeding module is used for identifying the control instruction transmitted by the signal processing module through a USB serial port and driving a mechanical arm to complete auxiliary feeding of corresponding food according to the control instruction; the SSVEP is induced through the stimulation of square flicker on a liquid crystal display screen, meanwhile, the SSVEP is effectively classified through an FBKCCA algorithm, electroencephalogram signals generated when a person with the upper limb disability watches flicker stimulation of different frequencies are processed, and the classification accuracy is higher.

The invention discloses a rectal cancer lymph node metastasis diagnosis method based on deep learning multi-modal CT, and the method comprises the steps: carrying out the data preprocessing of a rectal cancer multi-modal CT image, and extracting the image features of a cut 3D plain-scan CT image and a cut 3D enhanced CT image through a newly constructed Mlenet (multi-modal Lenet convolutional neural network) convolutional neural network; and splicing the feature maps to form a new feature map, inputting the new feature map into a new Mlenet convolutional neural network, and performing dichotomy prediction to obtain a dichotomy prediction result. According to the method, effective feature extraction can be carried out on the multi-modal CT image, and the accuracy of lymph node metastasis prediction is greatly improved.

The invention discloses a fault feature adaptive extraction method based on wavelet entropy and EEMD (ensemble empirical mode decomposition). The method comprises the following steps: performing intrinsic mode decomposition on collected mechanical product operation historical signals by using ensemble empirical mode decomposition; calculating the wavelet energy entropy of the intrinsic mode function component obtained by decomposition; extracting an intrinsic mode function component containing most complex information; wavelet threshold noise reduction is carried out on the extracted intrinsic mode function component; and recombining the intrinsic mode function components after noise reduction. According to the method, the intrinsic mode function component with most energy information is adaptively determined by using the wavelet energy entropy, so that adaptive screening of the intrinsic mode function can be completed without depending on experience and test, and the fault characteristic frequency band is quickly extracted.

The invention discloses an image super-resolution implementation algorithm based on an information filtering network. The algorithm comprises a feature extraction module, a plurality of superposed information filtering modules and a reconstruction module, wherein the feature extraction module extracts features from a low-resolution image firstly, then superposes a plurality of information filtering modules, extracts residual information step by step, and finally, the reconstruction module aggregates the obtained high-resolution residual feature representation to generate a residual image; andelement-level item-by-item addition operation is performed on the residual image and the up-sampled low-resolution image. According to the image super-resolution implementation algorithm based on theinformation filtering network, due to the fact that the number of filters of each layer is relatively small, and group convolution is used, the network algorithm has the advantage of being high in execution speed.

The invention discloses a college poor student precise subsidy model based on a long short-term memory (LSTM) neural network, which constructs the college poverty student precise subsidy model based on the LSTM neural network, analyzes the number of hidden layer neural units of the LSTM neural network through a data experiment, trains a relationship between an optimization algorithm and a povertystudent subsidy level recognition rate, and structure and parameter optimization is carried out on the college poor student accurate subsidy model based on the LSTM. The accurate subsidy model can identify the national college subsidy level according to the college poverty student consumption data, provides an intelligent quantification tool for accurate identification and classification of college poverty student subsidy, can reduce human interference factors, and has innovative significance for college student subsidy and learning assistance work.

The invention discloses a lower limb movement recognition method based on a pressure and acceleration sensor. The specific implementation steps of the method are as follows: firstly, the pressure sensor signal of the lower limb movement of the human body is collected in real time, and after preprocessing the pressure sensor signal, according to the pressure sensor data rising The edge and falling edge mark the start and end of the lower limb movement. When the rising edge of the pressure is detected, the three-axis acceleration signal of the acceleration sensor will be collected and stored. When the falling edge of the pressure is detected, the three-axis acceleration signal of the acceleration sensor will be collected. The three-axis signal of the acceleration sensor collected between the edge and the falling edge is called the acceleration signal segment. Then the frequency domain features and statistical features are extracted from the acceleration signal segment extracted in the previous step. After the features are extracted, the data dimensionality reduction is performed on the extracted features. Finally, the trained classifier is used to classify the feature data after dimension reduction, and the classification result of the action pattern is obtained.

A human body posture recognition method using a millimeter-wave radar comprises the steps that echo data of the millimeter-wave radar in the detection range of the millimeter-wave radar are acquired, and the echo data of the millimeter-wave radar comprise point cloud information; determining that a human body exists in the detection range according to the millimeter wave radar echo data; determining the mass center of the human body according to the point cloud information of the millimeter wave radar echo data; noise of the point cloud information is removed according to the mass center of the human body; and recognizing a human body posture according to the denoised point cloud information. The invention further provides a computer readable storage medium, privacy leakage is avoided, and the accuracy of human body posture recognition is improved.

The invention relates to an all-round fisheye image target detection method based on a cascaded neural network, which is used for solving the problem of target detection difficulty caused by serious target distortion of an all-round fisheye image, and comprises the following steps: constructing a Cascade RCNN all-round fisheye image target detection network model fusing deformable convolution anddeformable pooling; training the reconstructed model; carrying out model measurement. When the network model is constructed, an improved Resnet50 network is used as a sub-network for image feature extraction, a fixed convolution and fixed pooling layer from stage 3 to stage 5 is replaced with a deformable convolution and deformable pooling layer, the receptive field and the self-learning offset can be expanded according to the state of a deformation target, effective feature extraction is carried out, and the accuracy of image feature extraction is improved. Therefore, the distortion target iseffectively modeled. Besides, the cascaded neural network is adopted, so that positive and negative samples can be subjected to multi-stage filtering for multiple times, and the accuracy of target detection is greatly improved.

The invention discloses a no-reference image quality evaluation method based on an adversarial generative network. The method comprises the steps of firstly preprocessing an image to obtain a similargraph-SSIM _ MAP corresponding to a distortion graph, then training a neural network framework based on a densenet network, inputting the distortion graph into the trained network to obtain the similar graph of the distortion graph, and obtaining a corresponding quality score through the similar graph. Innovations are made for a generator and a loss function in a network. The method comprises thefollowing steps: firstly, in a generation network part, adopting a 60-layer densenet network framework; in the discrimination network part, adopting a simple classification network; and in the loss function part, adopting a mode of adding L1 norm loss to the cross entropy of a discriminator; and finally, iteratively training a relatively good generation network model, and generating a similar picture for an output picture, namely a distorted picture, through the network.

The invention discloses a no-reference image quality map generation method based on an adversarial generative network. According to the method, a U-net network framework with eight down-samples and eight up-samples is adopted in a generation network part; a classification network is adopted in the discrimination network part; a mode of adding L1 norm loss to the cross entropy of a discriminator isadopted in the loss function part; and finally, a generative network model is iteratively trained, a similar graph of the input distortion graph is acquired through the generative network model, anda corresponding quality score is obtained through the similar graph. The method has no reference quality evaluation. Quality evaluation is performed on the distorted image under the condition of no natural image by using the trained neural network framework. The problem of calculating the quality score of the similar graph with the weight problem is solved. Based on an adversarial generative network and U-net, graph-to-graph conversion and migration are realized more effectively. An experiment result has a good result in graph-to-graph implementation, and the simulated mass fraction and the real mass fraction have strong correlation and small errors.

The invention discloses a no-reference image quality evaluation method based on a convolutional neural network. The method comprises the steps: preprocessing a distortion graph and a natural graph to obtain a similar graph, and then constructing a neural network according to the distortion graph and the similar graph; on the basis of the confrontation generation concept of a GAN framework, in a generation network part, integrating the jump connection characteristic of a U-net framework and the denseblock structure characteristic of a densenet framework; in the network discrimination part, adopting a simple classification network; and finally training the constructed neural network. According to the method, the characteristics of the GAN network, the U-net network and the densenet network are respectively absorbed and combined, a more effective neural network is constructed, conversion and migration from graphs to graphs are more effectively realized, a better result is obtained in graph-to-graph realization, and the simulated quality score and the real quality score have strong correlation and smaller error.

The invention relates to a relation extraction method for an automatic knowledge graph construction system ; the method comprises the steps: coding and converting a text into word vectors, and preliminarily extracting text features; generating a syntactic dependency tree by using a syntactic dependency structure of the text, weighting each relation category to generate a weighted dependency adjacency matrix, and extracting syntactic dependency information in the text by using a graph convolutional neural network; synchronously, directly acting a multi-head attention mechanism on the encoded text to generate an attention matrix, and extracting information except syntactic dependency information of the text by using a graph convolutional neural network with the same structure; and finally, obtaining feature representations of the two entities and sentences, scoring all possible relation categories by using a feedforward neural network and a normalized exponential function, and selecting a relation with the highest score as a relation classification result. According to the method, information of different dimensions of the text can be fully acquired, and an excellent effect is achieved on a public data set for relation extraction.

The invention discloses a method for measuring enterprise management efficiency based on compound optimization analysis, comprising the following steps: A. collecting enterprise operation data, performing preprocessing on the collected data, removing irrelevant attributes, and eliminating relevant high-branch attributes; B. Cluster and classify the preprocessed data to obtain the enterprise operation data after clustering and classification; C. Extract data features from the data after clustering and classification to obtain the data after feature extraction; D. Finally, realize the enterprise operation data through comprehensive measurement Measures of management efficiency. The measurement method of the present invention is simple to operate, has high measurement efficiency, and improves enterprise management efficiency; among them, the data clustering and classification method adopted reduces the time spent on manual labeling of data sets, makes data classification more accurate, and improves data efficiency. Utilization rate; the data feature extraction method adopted can effectively extract the characteristics of enterprise data, which is convenient for subsequent comprehensive measurement.

The invention discloses a method for measuring enterprise management efficiency based on compound optimization analysis, comprising the following steps: A, collecting enterprise operation data, preprocessing the collected data, removing attributes irrelevant to the data, and removing related high branch attributes; B, clustering and classifying the pre-processed data to obtain the enterprise operation data after clustering and classifying; C, extracting data features from the data after clustering and classification to obtain the data after feature extraction; D, finally, measuring that management efficiency of the enterprise through the comprehensive measurement. The measuring method of the invention has the advantages of simple operation, high measuring efficiency and improved enterprisemanagement efficiency. Among them, the method of data clustering and classification reduces the time consuming to label the data set manually, which makes the classification of data more accurate andimproves the utilization rate of data. The adopted data feature extraction method can effectively extract the characteristics of enterprise data, which is convenient for the subsequent comprehensivemeasurement.

The present application relates to a Chinese named entity recognition method, comprising: obtaining a sentence to be recognized, inputting the sentence to be recognized into an embedding layer of a preset neural network, and outputting an output word vector of the sentence to be recognized; Synchronously input the preset bidirectional long-term short-term memory network and the preset empty convolution network to obtain the output matrix; input the output matrix into the preset segmented long-term short-term memory network to obtain multiple prediction sequences; use conditional random field algorithm An optimal sequence is selected from the predicted sequence, and the identified entity is obtained according to the optimal sequence. In this application, by synchronously adopting bidirectional long-term short-term memory network and preset hollow convolution network for feature transformation, features can be effectively extracted without relying on entity dictionaries, and recognition efficiency is improved.