62 results about "Fisher criterion" patented technology

The invention discloses a feature selecting method for pattern classification. The method includes on the basis of figuring out a unsupervised optimal discriminant vector by adopting fuzzy Fisher criterion as an objective function, figuring out each feature importance weight according to a value of each dimension of the vector, collating the features according to dimension of the weights, selecting subsets of the features through a given threshold, and accordingly realizing feature dimension reduction. According to the feature selecting method, sample classify information is not required to provide beforehand, the problem that feature selection is lack of separation measurement in a unsupervised mode is solved effectively, good dimension reduction performances are realized in UCI data set and fault diagnosis experiments, and the feature selecting method is highly practical.

Provided is a method for effectively segmenting a hyperspectral oil-spill image. The method comprises steps of: defining an initial level set function and other related functions; acquiring a new fitting item in combination with a Fisher criterion; constructing an edge stop function to obtain a new length item; performing improvement in combination with an end member extraction algorithm; introducing a level set regular item to prevent reinitialization of the level set function; minimizing an energy function to obtain an Euler-Lagrange equation; setting parameters; selecting a display band and an initial contour; displaying a segmentation result graph; calculating various segmentation precision evaluation indexes; comparing and evaluating the accuracy of the segmentation results. The method can classify a target area in a simulated hyperspectral image and a real hyperspectral image, and effectively segments the hyperspectral oil-spill image with boundary blur and noise, improves the segmentation accuracy of the hyperspectral image, obtain a more accurate classification effect, makes the parameter change more stable, makes the contour curve more accurate, obtains the continuous and closed boundary contour, and has higher precision of segmentation.

The invention relates to a Gabor human face recognizing method based on a simplified intelligent single-particle optimizing algorithm, comprising a simplified intelligent single-particle optimizing algorithm and a link of selecting a Gabor filter group by the simplified intelligent single-particle optimizing algorithm, wherein the simplified intelligent single-particle optimizing algorithm comprises the following steps of: carrying out optimizing search on a solution space of a problem function by a particle and respectively carrying out intelligent updating on all dimensionality components of the single particle during iteration; and the link of selecting the Gabor filter group by the simplified intelligent single-particle optimizing algorithm comprises the following steps of: constructing a particle structure according to the quantity of Gabor filters, determining a particle search range and using a Fisher criterion as a fitness function; during the human face recognition, carrying out characteristic extraction on a human face image conforming to the authentication to form a characteristic database; after a new human face image is input, carrying out the characteristic extraction on the human face image by the selected Gabor filter; and comparing the obtained characteristic vector with the characteristics in the database by a minimal adjacent classifier to judge whether the input image is contained in the database or not and a corresponding identity. The invention improves the recognition precision and the characteristic extracting pertinence.

The invention relates to a method for identifying the surface contamination condition of a DC line insulator based on a visible light image. The method comprises the steps of preprocessing an image by adopting the visible light image segmentation algorithm, establishing the relationship between the contamination level and the visible light image, and identifying the contamination condition. According to the embodiment of the invention, the method comprises the steps of acquiring the visible light image of an insulator; preprocessing the image: including segmenting the image, removing the complex background of the image, filtering out the interference information, and ensuring the accuracy of the subsequent feature extraction and the contamination identification. The method specifically comprises the steps of converting an RGB model into an HSV color model; extracting the image color statistic features of the insulator, and evaluating the superiority-inferiority of each statistic feature according to a Fisher criterion function; forming characteristic vectors by adopting statistic features in accordance with requirements and inputting the characteristic vectors into an SVM classifier for classification; and outputting a contamination level. The method is suitable for all insulators having obvious differences from contamination color features. Therefore, the contamination levels of insulators can be effectively identified and the classification accuracy is high. The method provides a new technical means for the identification of the contamination condition.

The invention discloses a human face identification method based on fisher low-rank matrix restoration. According to the method, the human face identification problem is subjected to modeling under a low-rank matrix restoration framework, and in addition, a fisher discriminating criterion is combined for carrying out regularized discriminating restriction. In the human face identification problem, label information of all training images are known, the class-by-class discriminating criterion is utilized for regularizing the representation base obtained through the low-rank matrix restoration through being enlightened by the fisher criterion, and the discriminating performance of the algorithm in the human face identification is improved. The human face identification method has the characteristics that most sparse noise can be effectively removed; when both training images and test images are damaged, the human face identification performance is obviously superior to other algorithms.

The invention discloses a Gabor filtering and deep neural network-based warp-knitted jacquard fabric defect detection method. The method comprises two parts including a training stage and a detectionstage. In the model training stage, a defect-free fabric image is adopted; a training sample set is obtained through image preprocessing; and then parameters in a deep neural network are trained. During detection, a detection sample set of a to-be-detected image is obtained; then a network output is obtained by adopting the deep neural network; and finally a defect detection result is obtained. AGabor filter is adopted for extracting texture features of a fabric; a cost function value is calculated according to a Fisher criterion; textures of all directions and sizes in the image can be extracted; and the detection accuracy is improved. Image features are automatically selected by adopting the deep neural network, so that the defect and deficiency of artificial feature selection in an existing robot vision-based defect detection method are overcome.

The invention provides an embedding manifold regression model based on a Fisher criterion. A method for the embedding manifold regression model comprises the following steps of performing initializing; expressing a training sample by using a matrix (img file= ' DDA0000627843280000011. TIF' wi= ' 581' he= ' 56'/) under the condition that a category label corresponding to xm is that l (xm) belongs to {1, 2, ..., c}; preprocessing the training sample: mapping the training sample to principal component analysis subspace; establishing a similar matrix; separately processing a within-class sample and an inter-class sample by using the Fisher criterion; calculating embedding subspace: defining a Dxd mapping matrix W= [Omega1...Omega d] under the condition that d is dimensionality of the sample after the sample is subjected to feature conversion; and finding out mapping subspace by solving a feature vector (img file= ' DDA0000627843280000013. TIF' wi=' 123' he=' 55' /) of a matrix (img file= ' DDA0000627843280000012. TIF' wi=' 205' he=' 58' /). A conversion mode of the sample from the original higher-dimensional space to lower-dimensional manifold space is yi= WTx i=F (x i), and matrix representation of the conversion mode is Y=WTX=F(X), Y= [y(1), ..., Y(M)]. On the premise that label information of the sample is sufficiently used, local geometric structures of the same samples are maintained before and after dimensionality reduction, and the similarity of the samples which are in different categories but are high in similarity in the original space is reduced after dimensionality reduction.

Based on combination between wavelet transform and a Fisher criterion, the invention provides an EEG feature extraction method specific to a P300 Speller Brain-machine interface, mainly comprising the steps of: configuring a wavelet transform matrix for an EEG data segment of a specific length according to a wavelet function and a wavelet decomposition level designated by a user; utilizing a wavelet transform matrix to map the EEG data segment to a wavelet domain, and utilizing the Fisher criterion to search for a projection axis which can distinguish different types of data to a maximum degree in the wavelet domain; utilizing the projection axis to extract a plurality of rows from the wavelet transform matrix to form a feature extraction matrix; mapping a corresponding EEG data segment to a feature vector by a feature extraction matrix of each channel; and splicing feature vectors on all channels to form a feature vector for each time of stimulation. The method stimulates a corresponding EEG data segment each time to calculate a feature vector, and can reduce times of stimulation repetition under the condition of meeting accuracy rate requirements so as to increases a character transfer rate.

The invention provides a redundancy removal feature selection method LLRFC (Locally Linear Representation Fisher Criterion) score+ based on LLRFC and correlation analysis. A DNA (Deoxyribonucleic Acid) microarray technology provides a new direction for clinic tumor diagnosis. Performance of gene expression data corresponding to different kinds of tumor is different; through the analysis on the tumor gene expression data, study personnel can realize the accurate recognition on the tumor and the tumor subtype in the molecular level; and important biological significance is realized on the diagnosis and the treatment of the tumor. The feature genes in LLRFC judging criterion descending sort gene expression data is used to be combined with the dynamic correlation analysis strategy for further eliminating redundant features; an LLRFC score+ algorithm is provided; and the optimum feature gene subset is selected. The feature selection method LLRFC score+ has the advantages that the classification precision of a classifier can be effectively improved; a sample data set does not need to meet the normal distribution; and the method is applicable to data in various distribution types. The feature selection method LLRFC score+ can help people to find the virulence gene of cancer, and the early-stage diagnosis, tumor staging and typing, prognosis treatment and the like of clinic tumor diseases are facilitated.

The invention discloses a method for calculating and identifying protein kinase phosphorylation specific sites, which is characterized in that, the method comprises the following steps: a) establishing a new protein sequence structure characterization method-amino acid three-dimensional character scoring based on the active ingredient analysis method; b) using the amino acid three-dimensional character scoring for characterizing the structure features of the protein kinase phosphorylation specific sites; c) using a Fisher criterion scoring method for selecting parameters which are closely related to the features of the protein kinase phosphorylation specific sites; and d) and establishing a protein kinase phosphorylation specific site identification model by a radial basis kernel support vector machine, carrying out the self-replacement verification respectively, and using the retaining 1/10 cross-verification and the external verification for verifying the predictive capability of the method. The method can be used for the identification of the protein kinase phosphorylation specific sites, explore the protein phosphorylation rules under physiological and pathological states, further elaborate the nature of life and the disease pathogenesis and provide the important support for developing new drugs.

The invention discloses a method for obtaining a character matching template, which comprises the following steps: A, acquiring a handwritten character sample of a character class; B, calculating a discrete coordinate sequence; C, adjusting the size and shape of the handwritten character sample and a coordinate value of a gravity center to obtain a regular coordinate sequence of the handwritten character sample; D, carrying out characteristic extraction to acquire a multi-dimensional characteristic vector of the handwritten character sample; E, selecting partial characteristic value from the multi-dimensional characteristic vector of the sample of the character class, and calculating a sample center of each character class to obtain a rough sorted template; and F, calculating a characteristic transformation matrix according to the Fisher criteria, carrying out characteristic transformation for the multi-dimensional characteristic vector of the sample of the character class, recalculating the sample center of each character class, and then carrying out iteration regulation to obtain a fine sorted template. The invention also discloses a system for obtaining the character matching template. The rough sorted template and the fine sorted template obtained by the method improve the accuracy of handwritten character recognition.

The invention belongs to the field of wireless positioning, and relates to a CSI system multipath classification method based on an improved random forest algorithm. The method aims at distinguishingtwo propagation modes of LOS and NLOS, and comprises the following steps: clustering the energy of all samples by using a K-means clustering algorithm according to the propagation characteristics of wireless signals in LOS and NLOS environments, and constructing a characteristic factor based on the K-means clustering algorithm; calculating inter-class scattering distances and intra-class scattering distances of all the samples, and obtaining an optimal feature combination according to a Fisher criterion; and training and testing different feature combinations by using a random forest algorithmbased on a C4.5 algorithm to complete multipath classification. The method has the advantages that the phenomenon that multipath classification is limited by an accurate threshold value is effectively avoided, and the requirements for high accuracy and low operand in a multipath classification application scene can be met.

The invention discloses an electrode optimization method based on a brain source imaging technology. An improved Fisher criterion is used for solving the F score of each collected electrode which is evenly distributed; a brain source imaging technology is used for carrying out inverse transformation on the obtained F score to the cortex to obtain corresponding brain source distribution; an electroencephalogram forward problem is solved, i.e., on a premise that a transfer matrix between the cortex and dandruff is known, and the estimation value of the F score of an unknown electrode is solved;and finally, the estimation value of the F score is subjected to descending sorting, and parts of electrodes with a highest score value are selected as an optimal electrode. By use of the method, theproblems that electrode optimization depends on experiences or an amount of experiment collection electrodes in a motor imagery task is large can be solved, and the functions that a small quantity ofelectrodes is used for pre-estimating the contribution degree of the unknown electrode and the electrode optimization is carried out can be realized.

The invention discloses an image classification method based on learning of a FISHER multi-level dictionary. The method includes the steps of constructing an image classification system, importing sample images, and utilizing a sparse representation matrix of an over-complete dictionary to conduct primary appropriate classification on each sample image; conducting learning at different levels on the images classified by the over-complete dictionary on the basis of the FISHER multi-level dictionary, and obtaining specific information and common information of each class of images; using a discriminant of a FISHER criterion principle to reinforce the distinguishing capacity of the dictionary at different levels to form an optimal image classification discrimination standard. The image classification method based on learning of the FISHER multi-level dictionary can better catch various images which have most distinctive features under comparison with other images, each image has the mostunique features, and the distinguishing efficiency and accuracy are improved.

The invention provides a voice recognition method and system based on Fisher hybrid features and a neural network, and relates to the technical field of voice recognition. The method comprises the following steps of firstly, obtaining a to-be-tested voice and a voice sample set including intelligent synthetic voice data and natural human voice database data, and then obtaining MFCC features and CQCC features of voice samples in the voice sample set; secondly, obtaining MFCC-CQCC hybrid features of the voice samples based on Fisher criteria, the MFCC features and the CQCC features; thirdly, obtaining a voice recognition model based on the hybrid features and the preset neural network; and finally, judging whether the to-be-tested voice is an intelligent synthetic voice or a natural human voice based on the voice recognition model. According to the voice recognition method and system, in voice feature selection, single features are not selected and the MFCC-CQCC hybrid features based onthe Fisher criteria are selected; the hybrid features organically combine the MFCC and CQCC features, so that the voice synthesized by multiple algorithms can be effectively recognized; and the neuralnetwork is trained by using the hybrid features to obtain the voice recognition model, so that the accuracy of the voice recognition model can be effectively improved.

The invention discloses a color image segmentation method based on coring fuzzy Fisher criterion clustering. The color characteristic of a color image is mapped into a high dimensional space by using the method, clustering is realized by taking a coring fuzzy Fisher criterion as a target function in the high dimensional space, and thus image segmentation is finished. The invention solves the limitation that the traditional image segmentation method is difficult to process linear quarantine data, improves the noise robust property, obtains higher image segmentation quality, and has relatively high practical value.

The invention provides a speech emotion recognition method based on spectral features and ELM. The method comprises the following steps of extracting basic characteristics of an original speech signal, wherein the basic characteristics comprise rhythm characteristics and tone quality characteristics; extracting a Mel frequency cepstrum coefficient (MFCC) and a cochlear filter cepstrum coefficient(CFCC) by using a Teager energy operator TEO algorithm, weighting the MFCC and the CFCC to obtain a teCMFCC characteristic, and fusing the teCMFCC characteristic with a basic characteristic value to construct a characteristic matrix; performing selective dimension reduction on the characteristics by using a Fisher criterion and correlation analysis, and reserving the individual characteristics ofthe speech signals; and establishing an ELM decision tree model of an extreme learning machine to finish speech emotion recognition and classification. Through the method provided by the invention, the nonlinear characteristics of the speech signals are emphasized, and good robustness is provided; the test is performed on the CASIA Chinese emotion corpus recorded by the automatization institute ofChinese academy of sciences so as to verify that the proposed speech emotion recognition algorithm based on the spectral characteristics and the ELM has good classification and recognition precisionon the Chinese speech signals.

The invention discloses a rolling bearing fault diagnosis method based on a full convolution auto-encoder and an optimized support vector machine, and belongs to the field of bearing fault diagnosis. Firstly, due to the fact that a traditional fault feature is difficult to extract and the generalization of the feature is weak, the invention provides a fault diagnosis method based on a full-convolution auto-encoder, and the full-convolution auto-encoder has the advantages of a convolutional neural network and an auto-encoder at the same time; a stacked full-convolution auto-encoder is adopted to automatically extract depth fault features from the bearing vibration signal frequency spectrum; then, the Fisher criterion is used for grading and sorting the extracted depth fault features, the criterion is based on the intra-class distance and the inter-class distance, and the fault features with high discrimination can be screened out; and finally, optimizing hyper-parameters of the SVM by adopting an improved doliolaria algorithm, and inputting the screened features into the optimized SVM to complete fault identification of the rolling bearing.

The invention provides an area partitioning and positioning method based on space partitioning. The method comprises the following steps that 1, after a user designates the subarea number, various optimal initial cluster center positions are determined through distance calculation, the within-class distance, the between-class distance and the reference point spacing limiting condition are introduced by integrating a K-means algorithm, a Fisher criterion and a self-organizing iterative analysis algorithm, and by means of multiple times of the cluster merging and splitting process, the area partitioning target is completed finally; and 2, subarea positioning is achieved by applying a machine learning technology in a semi-supervised mode. The invention further provides an area partitioning and positioning system based on space partitioning.

The invention discloses a radar HRRP target recognition method based on selective principal component analysis. The Fisher criterion is used to realize the selection of principal components with maximum separability among different targets; then the principal components with maximum separability among selected targets are used to realize the fast reconstruction of the original HRRP signal; and finally, radar HRRP target recognition is realized through the reconstruction error between the original HRRP signal and the reconstructed HRRP signal. The invention reduces the reconstruction time and computer operation resources, achieving the effect of further improving the target recognition rate; avoids the complex kernel function algorithm; uses simple Fisher selection criteria to effectively reduce the complexity of the algorithm and improve the real-time performance of the algorithm; and maintains the recognition effect equivalent to the former.

The invention discloses a porcelain insulator pollution grade detection method based on color characteristics. The porcelain insulator pollution grade detection method comprises transforming a color image channel to obtain an H component, then segmenting the H component by means of a two-dimensional minimum error method combined with morphological filtering to extract an insulator disk area, extracting seven characteristic values including the mean value, the maximum value, the minimum value, the range, the variance, the grayscale anisotropy and the gray scale entropy of six channels of the disk area, and screening out characteristics which have high classification ability by means of a Fisher criterion function as pollution grade discrimination characteristics, finally taking the discrimination characteristics of the training set as an input and taking the pollution grade as an output to train a mind evolutionary algorithm (MEA) optimized BP neural network, carrying out simulation prediction by means of test set data and judging the accuracy, and achieving non-contact on-line and high-efficiency detection of the insulator pollution grade. By employing the porcelain insulator pollution grade detection method based on the color characteristics, the problem that the existing pollution grade detection method cannot achieve on-line detection is solved.