49results about How to "Remove redundant features" patented technology

The invention discloses a brain tumor segmentation method based on a deep neural network and a multi-modal MRI image. The method includes steps: constructing the deep neural network, wherein the deep convolution neural network includes two three-layer convolution layers, a three-layer full connection, and a classification layer, an input layer corresponds to the multi-modal MRI image, and each node of an output layer corresponds to a tumor classification label; performing MRI image preprocessing; training a network model; and testing the model, performing normalization on a to-be-segmented tumor image sequence by employing image blocks of an MRI image sequence and mean values and standard deviations thereof in a training process, inputting the normalized image sequence to the deep neural network with the optimization network connection weight, obtaining node values of the classification layer, and obtaining the tumor classification of a to-be-segmented brain tumor image. According to the method, tumor abstract topological characteristic information in the multi-modal MRI image is mined and extracted by employing the deep neural network, and high segmentation accuracy and high segmentation precision can be guaranteed in brain tumor segmentation of the multi-modal MRI images.

The present invention claims a handwriting numeral recognition method based on a convolutional neural network and a support vector machine. The method organically combines a convolutional neural network model with a support vector machine model. A handwriting numeral recognition model combining the convolutional neural network with the support vector machine can deeply describe correlation between sample data and expected data and automatically learn an image feature from original data, has very good decision plane, and is very strong in discrimination capability of digital pattern classification. The handwriting numeral recognition method based on the convolutional neural network and the support vector machine, provided by the present invention, is simple, easy to implement, and very good in handwriting numeral recognition effect.

The invention discloses a face identification method which blends a global feature and a local feature based on an ICA. As a DCT can efficiently transit a high-dimensional face image into a low-dimensional space and reserves most of identifiable information of the image, the DCT is suitable for extracting the global feature of the image, while a Gabor wavelet transformation is suitable for extracting the local feature and the classification feature of the image, and the two features are widely applied to the face identification. Based on the two methods, we bring in the ICA technology to extract the independent Gabor feature and the independent DCT feature of the image, then efficiently blends the independent Gabor feature and the independent DCT feature to obtain a novel independent feature, enables the novel independent feature to have the local information of the Gabor feature and the global information of the DCT feature, efficiently reduces the dimension of a feature vector, and removes superfluous features. Finally, the blended independent feature is used for the SVM to realize the face classification identification.

The invention provides a pedestrian re-identification method and system, a computer readable storage medium. The pedestrian re-identification method comprises the following steps: obtaining a calibration data set, and training the calibration data set to form a segmentation model; acquiring a pedestrian image, and segmenting the background of the pedestrian image to obtain a foreground image and an environment image; extracting body-shaped key points of pedestrians in the foreground image containing the pedestrians, and segmenting the foreground image based on the body-shaped key points to form an ROI; extracting features of the foreground image and the ROI of the region of interest based on a feature extraction model to obtain global features and weighted features, and connecting the global features and the weighted features in series to form a multi-dimensional feature vector; and performing similarity comparison on the multi-dimensional feature vector and features extracted from thetarget pedestrian to determine whether the pedestrian is the target pedestrian. By removing background images of pedestrians captured under different cameras, redundant features during feature extraction are eliminated, recognition results of pedestrian re-recognition are only based on pure features, and the occurrence of false recognition is reduced.

The invention provides a method for constructing and training a human face identification feature extraction network. The method comprises the steps of constructing a feature extraction network and a metric learning dimension reduction network, wherein an output of the feature extraction network is an input of the metric learning dimension reduction network; training the feature extraction network based on all sample sets to output feature sets; screening the feature sets by utilizing semantic sampling to obtain a pure sample set; and training the metric learning dimension reduction network based on the pure sample set. Through a natural human face identification network constructed by the method, the feature representation capability can be improved, so that feature information in data can be fully mined and an original human face picture can be accurately identified.

The invention discloses a rotary mechanical health assessment method for a deep self-encoding network. The method comprises the steps of (1) vibration signal acquisition, (2) original feature extraction, (3) feature dimension reduction by using a deep auto-encoder (DAE) network, (4) feature selection, (5) health indicator construction by using an unsupervised SOM algorithm, and (6) health indicator evaluation by using a fusion evaluation criterion based on a genetic algorithm. According to the method, the advantages of the powerful feature extraction ability of deep learning are combined, deepself-encoding and minimum quantization error methods are combined. In addition, an evaluation criterion based on one metric often has a bias problem, and the invention provides the fusion evaluationcriterion based on the genetic algorithm. According to the method, the health state of rotary machinery can be accurately evaluated, the method can be widely applied to the health assessment of rotarymachinery in the fields of chemical engineering, metallurgy, electric power, aviation and the like, the dynamic process of performance degradation of these components can be accurately described, anda remaining life also can be predicted.

The embodiment of the invention discloses an expression recognition method, device and equipment for a video. The method comprises the steps: extracting a corresponding image sequence from any video comprising a face; extracting a face image in the video through face recognition, and carrying out a corresponding alignment operation as the input; carrying out the feature extraction comprising spatial features and time features through a pre-trained 3D convolution neural network, and generating a feature vector. The method integrates the prediction of the time-space domain of an image sequence,and achieves the more accurate expression recognition according to the feature vector.

The invention discloses a breast lump information extraction and classification method for a breast X-ray image. The method comprises the following steps of S1, inputting the breast X-ray image into four parallel convolutional neural networks; S2, extracting high-level semantic features of the breast X-ray image on the basis of the four parallel convolutional neural networks; and S3, performing multi-label multi-task learning network training on the extracted high-level semantic features, and obtaining classification information of breast lumps. According to the breast lump information extraction and classification method for the breast X-ray image, redundant features extracted by the convolutional neural networks can be effectively removed; four classification tasks can be mutually entangled and constrained and mutually promoted through a multi-label multi-task network; clear breast lump classification information is provided for doctors; and auxiliary diagnosis of breast lump relateddiseases is provided.

The invention discloses a multi-modal emotion recognition method and system fusing an attention mechanism and DMCCA. The method comprises the following steps of: respectively extracting electroencephalogram signal features, peripheral physiological signal features and expression features from preprocessed electroencephalogram signals, peripheral physiological signals and facial expression videos; extracting electroencephalogram emotion features, peripheral physiological emotion features and expression emotion features with discriminability by using the attention mechanism; using the DMCCA method for the electroencephalogram emotion features, the peripheral physiological emotion features and the expression emotion features to obtain electroencephalogram-peripheral physiological-expression multi-modal emotion features; and classifying and identifying the multi-modal emotion features by using a classifier. According to the method, the attention mechanism is adopted to selectively focus on the features with higher emotion discriminability in each mode, and the DMCCA is used to fully utilize the correlation and complementarity between the emotion features of different modes, so that the accuracy and robustness of emotion recognition can be effectively improved.

The invention discloses a novel efficient power quality disturbance image feature extraction and recognition method. The method comprises the following steps: converting an electric energy quality signal into a gray level image, enhancing disturbance characteristics by using three methods of gamma correction, edge detection and peak-valley detection to obtain a binary image, and extracting nine characteristics of area, Euler number, angle second moment, contrast ratio, correlation, mean value, variance, inverse difference moment and entropy to construct an original characteristic set; carryingout sorting on the basis of the feature Gini importance degree, and determining the feature with the maximum influence on classification; and comprehensively considering the classification precisionand efficiency, determining the number of trees in the random forest, and constructing a random forest classifier by using the optimal feature subset to identify the power quality disturbance signal.According to the invention, 8 types of common power quality disturbance signals of voltage sag, voltage sag, voltage interruption, flickering, transient oscillation, harmonic waves, voltage cutting marks and voltage peaks under different noise environments can be identified efficiently and accurately, and the feature extraction efficiency of the disturbance signals is improved.

The invention provides a retrieval method, comprising the steps of: representing a database to be retrieved as a document_keyword matrix, wherein the number of rows of the document_keyword matrix is equal to the number n of documents, and the number of columns of the document_keyword matrix is equal to the number m of keywords; generating a target matrix to represent the improved database to be retrieved, wherein the generation process comprises the following procedures of: transposing the document_keyword matrix to form a keyword document matrix, and decomposing the keyword_document matrix into the product of a keyword vector matrix, a diagonal matrix and a document vector matrix by a singular value decomposition algorithm; and selecting the keyword vector matrix and multiplying the document keyword matrix by the keyword vector matrix to set up the target matrix; and retrieving in the improved database to be retrieved which is represented by the target matrix. By using the retrieval method provided by the invention, the retrieval speed and the efficiency are greatly improved.

The invention discloses a rotating machine fault diagnosis method based on wavelet packet decomposition. The method comprises the following steps: 1) collecting vibration signals of a rotating machinein a normal state and a fault state; 2) selecting a wavelet basis function for fault feature extraction; 3) according to the selected wavelet basis function, obtaining sub-signals of different frequency bands of the vibration signal through wavelet packet decomposition; 4) calculating a fuzzy entropy value of the sub-signal to obtain a fault feature vector; 5) performing feature importance sorting according to the correlation, and selecting a set number of fault feature vectors with top sorting results according to the sorting results; 6) using a classifier to construct a fault diagnosis model, dividing the selected fault feature vector and the category label into a training set and a test set, and using the training set as the input of the model to train the model; and 7) inputting the test set into the fault diagnosis model to obtain a fault diagnosis result. According to the method, high-quality fault features can be effectively extracted, and the accuracy of fault diagnosis is improved.

The invention relates to a detection, recognition and classification method for a J wave, specifically a J wave detection and classification method based on correlation analysis characteristic selection. The method comprises the following steps: carrying out the preprocessing of an ECG signal, and carrying out the denoising and segmenting of the signal; carrying out three-layer wavelet packet decomposition of the segmented signal, carrying out the analysis of a third-layer coefficient, and calculating the area between a research segment and a base line; calculating the second-order, third-order and fourth-order cumulant characteristics and energy characteristics of the third-layer coefficient; carrying out the correlation analysis of the cumulant characteristics and energy characteristics, and carrying out the characteristic selection according to a classification effect; employing the selected characteristics as the input of an SVM (support vector machine) classifier, carrying out the classification and recognition of a normal signal and a J-wave signal, and successfully detecting the J-wave signal. According to the invention, a correlation analysis characteristic selection method is used for characteristic screening, thereby removing the redundant characteristics and improving the classification accuracy.

The invention provides a network traffic abnormal behavior identification method based on an autocoder. The invention belongs to the crossing technical field of machine learning and information security combination. The category distribution of normal traffic data and abnormal traffic data in traffic data is balanced by using a comprehensive minority oversampling method, and an autocoder is combined, so that nonlinear structure information can be effectively extracted from mass data, and abnormal behaviors in network traffic can be identified.

The invention discloses a high-voltage circuit breaker fault intelligent diagnosis method based on an improved fuzzy Petri network. The method comprises the following steps of (1), through online monitoring, acquiring a plurality of sets of data samples in normal operation and faulted operation of the high-voltage circuit breaker, and extracting characteristic quantities in various signals; (2), based on a rough set theory, performing continuous characteristic quantity discretization and decision table reduction by means of an improved greedy algorithm for eliminating redundant characteristic quantities, and simplifying a fault diagnosis rule; (3), according to the simplified diagnosis rule, setting a corresponding database and transition, establishing a fuzzy Petri network reasoning model, and obtaining a corresponding MYCIN reasoning equation; and (4), acquiring testing data, preprocessing the testing data, inputting the testing data into the MYCIN equation for performing reasoning, and obtaining a fault conclusion. The high-voltage circuit breaker fault intelligent diagnosis method performs processing for aiming at low accuracy of the sampling data and improves fault diagnosis efficiency through equation reasoning. Furthermore the high-voltage circuit breaker fault intelligent diagnosis method can promote development of intelligent power grid technology and improves reliability and stability of a power system.

The invention discloses a self-adaptive robust constrained maximum variance mapping (CMVM) characteristic dimensionality reduction and extraction method for the diversified image retrieval of plant leaves. On the basis of research on the characteristic extraction of image manifold and selection level, by adoption of a CMVM semi-supervised manifold dimensionality reduction method, the discrimination of positive class local sub-concepts can be kept, and the discrimination of positive and negative classes namely concepts is strengthened. By the invention, a de-noising method and a CMVM strengthening positive local keeping algorithm are provided for keeping the discrimination of the sub-concepts; a linear approximation method is provided for solving the problem of outer point learning of a CMVM sample; an ordered layer maximum interval correlation evaluation function of diversified retrieval is provided for selecting CMVM manifold functions and estimating image intrinsic dimensionality; and a maximum difference intrinsic characteristic method for mining and discriminating positive intra-class sub-concepts from CMVM characteristics is provided for clustering diversified learning, and the diversity of plant image retrieval is improved.

The invention discloses a sparse joint model target tracking method based on a self-adaptive selection mechanism. When a sparse judgment model is constructed, more discriminatory features are extracted by use of a feature selection mechanism, confidence value measurement is taken as a constraint, and the target and the background can be better distinguished; when a sparse generation model is constructed, in combination with L1 regularization and PCA subspace reconstitution concept, the target not only reserves sufficient appearance information, but also can effectively resist outlier disturbance, and an iterative algorithm combining linear regression and soft threshold operators is proposed for the minimum solution of a target function. Compared with a conventional multiplicative combined mechanism, the self-adaptive selection mechanism based on the Euclidean distance is proposed, the deviation is calculated by comparing the difference between prediction results of the two models and the tracking result of the previous frame, whether the models degenerate is judged, and the more reasonable joint model evaluation function is constructed to improve the tracking accuracy.

The invention is suitable for the technical field of computer application, and provides an intrusion detection method and device, and the method comprises the steps: carrying out the preliminary dimension reduction of a data subset of each intrusion type through an information entropy theory, obtaining a feature point corresponding to the intrusion type, and carrying out the construction to obtaina feature topology corresponding to each intrusion type; training a preset intrusion detection model based on the harmonic function, and adjusting parameters for optimizing an ant colony algorithm and parameters of a classifier based on a support vector machine in the intrusion detection model; and obtaining a feature subset according to the parameters of the intrusion detection model and ant movement, and determining a target feature subset solved by the ant colony after a preset number of training times so as to detect an intrusion type existing in the to-be-detected object through the target feature subset. Parameters of the ant colony and the support vector machine are trained, optimized and improved according to a ten-fold cross validation method, redundant features in a data set areremoved, and the detection performance of the intrusion detection method is improved.

The invention discloses an engine life prediction method, a storage medium and computing equipment, and the method comprises the steps: removing the information redundancy features in engine state data according to the correlation degree between the features, and removing the features with the small correlation degree with a prediction target according to the correlation between the features and the prediction target; randomly selecting a sampling sample set, randomly selecting a feature subspace on the random sampling sample set, establishing a decision regression tree on the obtained randomsampling subspace, and obtaining a life prediction result under a corresponding feature combination by the decision trees on different random sampling subspaces; constructing an MLP model structure and a loss function, and obtaining MLP model parameters through Adam algorithm learning; and integrating prediction results of the decision trees based on the trained MLP model to obtain the remaining service life of the engine. According to the method, prediction values of decision trees in different random sampling subspaces are integrated through a learning method, the prediction accuracy and reliability are improved, and a basis is provided for maintenance and fault prediction of the aero-engine.

The invention provides a bearing health state monitoring method based on a convolutional auto-encoder, and belongs to the field of bearing fault prediction and health management. The method comprises the following steps: firstly, acquiring a full-life-cycle digital vibration signal and a health state marking value of a brand new bearing; extracting intrinsic mode component statistic characteristics of the digital vibration signals and depth characteristics learned by using a convolutional auto-encoder, splicing and screening the two characteristics, inputting the screened characteristics into a full-connection regression network for regression training, and finally obtaining a health state curve graph of the bearing in the full life cycle; then, obtaining a current health state curve graph of the same type of to-be-monitored bearing; and comparing the two curve graphs to obtain a health state monitoring result of the to-be-monitored bearing. According to the method, the integrity of the bearing vibration signal features is improved through the convolution self-encoder, redundant features are removed by using a feature sorting and feature selection method, and the relatively accurate bearing health state can be obtained.

The invention discloses a power grid transient stability evaluation method based on an adaptive differential evolution algorithm and ELM, and the method comprises the steps: obtaining disturbed dynamic data and disturbed steady-state data of a power grid simulation disturbed track, and constructing a sample set; optimizing the extreme learning machine by adopting an adaptive differential evolution algorithm comprising an improved mutation strategy and an optimal particle local optimization mechanism; training the optimized extreme learning machine by adopting the sample set to obtain a transient stability evaluation model; and according to the transient stability evaluation model, carrying out the quick stability judgment of transient change after power grid disturbance. By establishing analysis models of different fault disturbance scenes and stability relationships, relationships between historical change trends of different positions and different monitoring quantities and system stability are determined, transient stability hierarchical key features are extracted, and meanwhile, an ELM transient stability evaluation model is optimized based on an adaptive differential evolution algorithm, so the rapid stability judgment of transient change after power grid disturbance is realized.

The invention discloses an attention mechanism-based short-time single-lead electrocardiosignal atrial fibrillation automatic detection system, which is characterized by comprising a data sampling module, a preprocessing module, an atrial fibrillation automatic detection module and an optimization training module. According to the invention, the adaptive attention module is added, so that characteristics of a large number of electrocardiosignals can be accurately extracted, and redundant characteristics are removed; and meanwhile, by emphasizing and suppressing information, parameters in a network structure can be continuously updated and adjusted to pay attention to and retain some important electrocardiosignal characteristics. In addition, the bidirectional time domain sampling module is cascaded with the one-dimensional dense connection network, so that on one hand, more information of tiny fine-grained change related to the short-time single-lead atrial fibrillation signal can be extracted; on the other hand, the time domain features of the electrocardiosignals can be fully considered, the difference and complementarity between the atrial fibrillation signals are better concerned, and higher classification accuracy and specificity are obtained.

The invention provides a tumor image focus area prediction analysis method and system and terminal equipment, and belongs to the field of deep learning. The method comprises the steps of collection of image data, diagnosis text and medical history data and prediction analysis of a tumor focus area, and specifically comprises the steps of preprocessing the collected data, and extracting image features through a constructed fusion weighted extraction network model; constructing a one-dimensional vector by using the medical history characteristics of a patient, namely age, gender, Karnofsky performance state, apparent tumor growth speed and function deterioration speed, and extracting the medical history characteristics by using a constructed text characteristic extraction network model of a dynamic convolution kernel; and after text features obtained by using the CBOW network model are fused with medical history features and image features, performing focus area prediction analysis through the constructed double-layer weighted prediction analysis network model. The method can significantly improve the classification prediction effect of the tumor image.

The invention relates to a neural structure corresponding learning cross-domain emotion classification method for improving feature selection, and belongs to the field of natural language processing.The method comprises the following steps: firstly, selecting two different fields in an Amazon comment data set as a source domain and a target domain, preprocessing source domain and target domain data to obtain text contents of the source domain and the target domain, secondly, carrying out word form restoration on a text, eliminating redundant features, and carrying out vectorization processingon the text to obtain initial features of the text; screening out pivot features through a chi-square test feature selection method to serve as pivot features in cross-domain tasks, and the rest features being non-pivot features; performing pivot feature prediction on the non-pivot features in the two fields through neural structure corresponding learning by utilizing the obtained pivot featuresto obtain feature migration; and training a logistic classifier by using the initial features and the migration features of the source domain text, and testing by using the text features and the migration features of the target domain to obtain a classification result of the target domain.

The invention discloses a vehicle detection method and device based on an attention mechanism and feature weighted fusion. The method comprises the following steps: preprocessing a to-be-detected image; inputting the preprocessed to-be-detected image into a pre-trained vehicle detection model; generating a channel attention feature map and a space attention feature map by using a vehicle detection model, based on the preprocessed to-be-detected image, and by adopting a channel and space two-dimensional attention mechanism; performing differentiated feature fusion based on the channel attention feature map and the space attention feature map through a weighted bidirectional feature fusion network, and obtaining fusion features; based on the fusion features, obtaining a detection result containing vehicle position and size information. Compared with the existing vehicle detection technology, the method has the advantages that the parameter quantity is reduced and the detection precision is improved while the relatively high detection speed is maintained, and particularly, the detection effect on small-scale vehicles is remarkable.

The invention discloses a malicious software detection method based on mixing of an improved naive Bayes algorithm and a gated loop unit, and belongs to the field of software detection. A traditional Android defense mechanism is difficult to deal with rapid increase of the number and types of malicious software. The malicious software detection method based on mixing of an improved naive Bayesian algorithm and a gating loop unit comprises the steps that a to-be-detected software sample set file is decompiled through apktool, a decompilation resource file of an application program is obtained, a feature set is extracted from the decompilation resource file, the extracted features are geometrically sorted from low to high according to the number of times of use, the features with high frequency are selected and combined into a feature set, and the feature set is quantized; and a gating loop unit is employed to process the features with time sequence variation to detect the dynamic features. According to the method, the malicious software using the confusion technology can be effectively detected, and the detection accuracy is improved.

The invention discloses a self-adaptive robust constraint maximum variation mapping (CMVM) feature dimension reduction method for image retrieval of plant laminae. According to the self-adaptive robust CMVM feature dimension reduction method, study is performed from the feature extraction and selective levels of an image manifold, and the capacity of keeping the distinguishability of positive type local 'sub-concepts' and the capacity of enhancing the distinguishability of 'concepts' of positive and negative types are realized by a CMVM semi-supervised manifold dimension reduction method to provide effective service for the diversified image retrieval; according to the practical application of the image retrieval, and aiming at basic problems of CMVM, the invention provides a method for removing noise points; the problem of learning of sample exterior points of CMVM is solved by a linear approximation method; and the selection of CMVM manifold parameters and the estimation of the intrinsic dimension of an image are performed by designing an 'ordered' level maximum interval relevance evaluation function of the diversified retrieval, so that a self-adaptive robust CMVM algorithm for the diversified time retrieval is provided. By the self-adaptive robust CMVM feature dimension reduction method, redundant characteristics are removed, and the retrieval efficiency is improved.

The embodiment of the invention relates to the field of software defect prediction, and discloses a data processing method and device based on principal component analysis, and a computer readable storage medium, and the method comprises the steps: carrying out the dimension reduction of initial sample data, and obtaining the sample data of a preset dimension; acquiring a plurality of features ofthe sample data, and calculating the relevancy between each feature and a preset category, the preset category being one of a plurality of categories of the sample data; and removing the features of which the relevancy is less than the preset relevancy in the plurality of features, and taking the remaining features as identification features of the sample data. According to the data processing method and device based on principal component analysis and the computer readable storage medium provided by the invention, redundant features in the sample data can be removed, and the sample data withhigh discrimination is obtained, so that the prediction efficiency is improved.

The invention discloses a micro-seismic signal identification method based on a quasi-optimal Gaussian kernel multi-classification support vector machine, and belongs to the field of machine learningand data mining. The method comprises the following steps: firstly, dividing micro-seismic data according to channels, and performing data format conversion; secondly, performing feature extraction oneach piece of channel data according to a mean value and a variance, combining all channels of the same sample to form a new feature, and performing feature selection on the synthesized data by utilizing an optimal Gaussian kernel-like multi-classification support vector machine to generate a dimensionality-reduced unbalanced training sample set; thirdly, determining an under-sampling rate according to the non-equilibrium rate of the training sample, and carrying out under-sampling on the large class of samples; and finally, a multi-classification support vector machine is adopted to construct a microseism signal classifier after dimension reduction. According to the method, the influence of redundant features on classification can be effectively reduced; double dimensionality reduction is carried out on the channel characteristics and the combined characteristics, so that the microseismic signal dimensionality is effectively reduced, the accuracy and timeliness of a microseismic signal classifier are improved, and the accuracy of rock burst disaster early warning is improved.

The invention discloses a face identification method which blends a global feature and a local feature based on an ICA. As a DCT can efficiently transit a high-dimensional face image into a low-dimensional space and reserves most of identifiable information of the image, the DCT is suitable for extracting the global feature of the image, while a Gabor wavelet transformation is suitable for extracting the local feature and the classification feature of the image, and the two features are widely applied to the face identification. Based on the two methods, we bring in the ICA technology to extract the independent Gabor feature and the independent DCT feature of the image, then efficiently blends the independent Gabor feature and the independent DCT feature to obtain a novel independent feature, enables the novel independent feature to have the local information of the Gabor feature and the global information of the DCT feature, efficiently reduces the dimension of a feature vector, and removes superfluous features. Finally, the blended independent feature is used for the SVM to realize the face classification identification.