Classification method and device based on feature crossing and fusion feature selection

Abstract

The invention relates to a classification method and device based on feature crossing and fusion feature selection. The method comprises the following steps: constructing a data set comprising a plurality of pieces of voice data and labels; extracting feature data from each piece of voice data and constructing a spliced two-dimensional matrix according to the feature data; sequentially performing first fusion feature selection, adaptive feature crossing and second fusion feature selection on the spliced two-dimensional matrix according to the labels to obtain to-be-classified data, and dividing the to-be-classified data into a training set and a test set; the self-adaptive feature crossing is characterized in that if a linear correlation coefficient between features is greater than a threshold value, nonlinear transformation is carried out and then feature crossing is carried out, otherwise, feature crossing is directly carried out; inputting the training set and the label into a classification learning device to obtain a trained classification learning device; and inputting the test set into the trained classification learner to obtain a classification result. The device realizes the method. According to the method, two times of fusion feature selection are used, the problem of dimension explosion during feature crossing is solved, and good classification robustness is achieved.

Description

Technical Field

[0001] The present invention belongs to the field of machine learning, and in particular relates to a classification method and device based on feature crossover and fusion feature selection. Background Art

[0002] Time series refers to a type of signal collected over time. This type of signal has a time order characteristic and is a data point sorted by time. Time series analysis includes both the diagnosis of past data and the prediction of future data. Many applications have made important contributions to the development of time series. Typical applications include: medicine, weather, economics, and astronomy. Typical time series analysis methods include: statistical methods, machine learning methods, and integrated methods. In applications that use machine learning for classification, it has significant advantages, especially when processing time series data. This type of data is widely used in many applications such as energy, finance, and health, such as power load, stock price trends, electrocardiograms, and voice signals. Through time series classification, future trends can be predicted, abnormal situations can be detected, and more informed decisions can be made. Speech classification, as a category learning of the auditory channel, can help individuals quickly identify and judge sound stimuli, and has special survival significance.

[0003] Which feature extraction and feature selection is used is an important factor affecting the final effect. Therefore, exploring new feature extraction and feature selection has long been a research hotspot in the field of machine learning. In terms of feature extraction, in order to extract deep features in the data, a secondary feature extraction method is often used, including: first extracting features and then performing feature crossover on the extracted features, or after extracting features, performing secondary extraction, dimension increase or dimension reduction operations on the features. The feature crossover includes explicit feature crossover and implicit feature crossover. Implicit feature crossover usually uses a deep neural network to implement feature crossover, which is a highly complex method. Explicit feature crossover is usually based on linear transformations of existing features, such as inner products, Cartesian products, etc. This type of feature crossover has poor fit for nonlinear models. Only two types of features can be crossovered using a fixed feature crossover method at a time. It is impossible to adaptively select the best crossover method according to the characteristics of existing features, and the original existing features are not retained after crossover. In terms of feature selection, although a variety of feature selection methods based on different principles have been proposed, a single feature selection method is often used in the prior art. In different scenarios, it is not always possible to use the most appropriate selection for the current feature to be selected, so the best effect is often not achieved. Summary of the invention

[0004] The purpose of the present invention is to propose a classification method and device based on feature crossing and fusion feature selection in view of the defects of the existing display feature crossing method, such as difficulty in fitting nonlinear models, limited number of feature types of each crossing, poor adaptability, failure to retain original features, and low classification accuracy caused by single feature selection. The method first collects the voice signal of the subject, extracts multiple feature data using the collected voice data, performs a first fusion feature selection on the multiple feature data after all are spliced ​​to obtain first selected data, performs adaptive feature crossing on the first selected data to obtain crossed feature data, and the data to be classified obtained after the second fusion feature selection on the crossed feature data is used for machine learning, and finally obtains the ability to classify the input voice data through machine learning; the classification device is used to realize classification based on feature crossing and fusion feature selection.

[0005] In order to achieve the above object, the present invention adopts the following technical solutions:

[0006] As a first aspect of the present invention, a classification device based on feature crossover and fusion feature selection is proposed, comprising a speech acquisition module, a feature extraction module, a first fusion feature selection module, an adaptive feature crossover module, a second fusion feature selection module and a model training module;

[0007] The speech acquisition module is connected to the feature extraction module, the feature extraction module is connected to the first fusion feature selection module, the first fusion feature selection module is connected to the adaptive feature cross module, the adaptive feature cross module is connected to the second fusion feature selection module, and the second fusion feature selection module is connected to the model training module;

[0008] The speech acquisition module performs digital-to-analog conversion of the speech signal and transmits the resultant speech data to the feature extraction module; the feature extraction module extracts multiple features from the speech data and transmits the obtained multiple feature data to the first fusion feature selection module; the first fusion feature selection module selects a predetermined number of features from the feature data and transmits the obtained first selected data to the adaptive feature crossover module; the adaptive feature crossover module performs adaptive feature crossover on the first selected data and transmits the obtained crossover feature data to the second fusion feature selection module; the second fusion feature selection module selects a predetermined number of features from the crossover feature data and transmits the obtained data to be classified to the model training module; the model training module performs model training using the above data to be classified and finally derives the classification model with the highest training accuracy;

[0009] The voice collection module includes a microphone and an analog-to-digital converter;

[0010] The microphone in the voice collection module is used to collect voice, and the voice collected by the microphone is converted into voice data by an analog-to-digital converter and passed to the feature extraction module;

[0011] The feature extraction module includes a wavelet feature extraction unit, a filter feature extraction unit, and an entropy feature extraction unit;

[0012] The number of the wavelet feature extraction units is W, and W is greater than or equal to 2;

[0013] The number of the filtering feature extraction units is F, and F is greater than or equal to 1;

[0014] The number of the entropy feature extraction units is E, and E is greater than or equal to 1;

[0015] The feature extraction module receives the speech data, and the three feature extraction units run in parallel to extract feature data of W groups of wavelet features, F groups of filter features and E groups of entropy features respectively;

[0016] The feature data of the W groups of wavelet features, the F groups of filter features and the E groups of entropy features are transmitted to a first fusion feature selection module;

[0017] The first fusion feature selection module includes a first average MIC calculation unit, a first feature selection unit and a first feature supplement unit;

[0018] The number of the first feature selection units is C1, and C1 is greater than or equal to 2;

[0019] The first fusion feature selection module receives the feature data, uses the first average MIC calculation unit to calculate the average MIC of the feature and the label from the doctor's judgment, the feature and other features, and according to the MIC, the C1 first feature selection units are operated in parallel to select the corresponding number of features of the C1 category, and the first feature supplementation unit supplements the features according to the difference between the number of the C1 category features after the union and the number of features to be retained, obtains the data after the first selection and passes it to the adaptive feature crossover module;

[0020] The adaptive feature crossover module includes a linear correlation coefficient calculation unit, a nonlinear transformation unit, and a Kronecker product unit;

[0021] The adaptive feature crossover module receives the data after the first selection, calculates the average linear correlation coefficient between features using the linear correlation coefficient calculation unit, and determines whether the linear correlation coefficient is greater than or equal to the threshold value S. If not, that is, less than the threshold value S, the data after the first selection is passed as input data to the Kronecker product unit; if so, that is, greater than or equal to the threshold value S, the data after the first selection is passed through the nonlinear transformation unit and then passed as input data to the Kronecker product unit. The Kronecker product unit calculates the Kronecker product of the input data and the data after the first selection, obtains the crossover feature data and passes it to the second feature selection module;

[0022] The value range of the threshold S is 0≤S≤1;

[0023] The second feature selection module includes a second average MIC calculation unit, a second feature selection unit and a second feature supplement unit;

[0024] The number of the first feature selection units is C2, and C2 is greater than or equal to 2;

[0025] The second feature selection module receives the crossover feature data, calculates the average MIC of the feature and the label, and the feature and other features using the second average MIC calculation unit, and according to the MIC, the C2 second feature selection units run in parallel to select the corresponding number of features of the C2 category, and the second feature supplementation unit supplements the features according to the difference between the number of the C2 category features after the union and the number of features to be retained, obtains the data to be classified and passes it to the model training module;

[0026] The model training module includes several classification models;

[0027] The model training module receives the data to be classified and puts it into several classification models for training, and takes the model with the highest training accuracy as the output of the model training module.

[0028] As a second aspect of the present invention, a classification method based on feature crossover and fusion feature selection is proposed, comprising the following steps:

[0029] S1, converting the collected voice into digital form to obtain voice data and adding a label to each voice data according to the doctor's diagnosis result;

[0030] The number of voice data is X; the value range of X is 2≤X;

[0031] S2, extracting several types of features from the speech data;

[0032] The number of feature types is N; the value range of N is 1≤N;

[0033] The nth feature of each voice data has Yn feature data and the number of types of the feature data is the same as the number of feature types; wherein n=1, 2, ..., N, N is the number of feature types; the total number of the features of the several types is

[0034] S3, concatenating the feature data corresponding to each voice data, obtaining a concatenated vector and then expressing it as a concatenated two-dimensional matrix;

[0035] The length of the concatenated vector is The spliced ​​two-dimensional matrix is ​​a two-dimensional matrix with X rows and Y columns, including feature data of all X voice data, each row corresponds to a voice data, and each column corresponds to a feature;

[0036] S4, performing the first fusion feature selection on all features in the spliced ​​two-dimensional matrix according to the labels to obtain the data after the first selection;

[0037] The first fusion feature selection is specifically as follows:

[0038] S41, calculating the maximum mutual information coefficient between the feature and the label in the feature data and the average maximum information coefficient between the feature and other features;

[0039] The feature is the yth feature, where y=1, 2, ..., Y; the maximum mutual information coefficient with the label is MIC1 y ;MIC1 y The value range is 0≤MIC1 y ≤1; the average maximum information coefficient with other features is MIC2 y ;MIC2 y The value range is 0≤MIC2 y ≤1;

[0040] S42, calculating the mean of the maximum mutual information coefficients of all features and labels and the mean of the average maximum information coefficients of the feature and other features;

[0041] The mean of the maximum mutual information coefficient with the label The value range is The mean of the information coefficients maximized by averaging the other features The value range is

[0042] S43, according to S42 and Use the corresponding feature selection method to select the feature data to be retained and obtain the selected features;

[0043] The S43 is specifically:

[0044] S431, using different feature selection methods to select corresponding numbers of features according to the number of feature data that you want to retain;

[0045] The number of feature data to be retained is M; the value range of M is 1≤M≤Y; the number of feature selection methods is C1, and at least one method that does not consider the correlation between features and one method that considers the correlation between features; the value range of C1 is 2≤C1; the corresponding number is and indivual;

[0046] S432, combining the features obtained by using different feature selection methods in S431 to obtain selected features to be supplemented;

[0047] The number of selected features to be supplemented is Mm, where m represents the number of features repeatedly selected by different feature selection methods;

[0048] S433, supplement the features repeatedly selected in S432 to obtain selected features, specifically:

[0049] like Select m non-repeating features after removing the features in S431 under the method that does not consider the correlation between features in S431 for supplementation; otherwise select m non-repeating features after removing the features in S431 under the method that considers the correlation between features in S431 for supplementation;

[0050] S44, selecting the feature data in S3 according to the selected feature to obtain the first selected data;

[0051] The number of rows of the data after the first selection is X, the number of columns is M, each row corresponds to a piece of voice data, and each column corresponds to a feature;

[0052] S5, performing adaptive feature crossover on the first selected data in S44 to obtain crossover feature data, specifically:

[0053] S51, calculating the linear correlation coefficient between every two columns of feature data of the first selected data in S44 and then calculating the mean of all linear correlation coefficients;

[0054] The linear correlation coefficient r ij ,i,j=1,2,...,M,j≠i;r ij The value range is 0≤r ij ≤1; the mean of the linear correlation coefficient The value range is

[0055] S52, if S51 If the value is less than the threshold value S, the Kronecker product of each row of data in S34 with itself is calculated respectively; otherwise, the result of each row of data passing through the nonlinear function is calculated first, and then the Kronecker product of the result and the row of data is calculated;

[0056] The threshold S has a value range of 0≤S≤1; the Kronecker product is a square matrix with a dimension of M in both cases;

[0057] S53, arranging the square matrix elements corresponding to each row of data obtained in S52 into row vectors, so that the characteristic data corresponding to all physiological data can be represented by a two-dimensional matrix, which is called cross characteristic data;

[0058] The length of the row vector is M 2 The number of rows of the cross-feature data is X, and the number of columns is M 2 , each row corresponds to a voice data, and each column corresponds to a cross feature;

[0059] S6, performing a second fusion feature selection on the cross features of the cross feature data obtained in S53 according to the labels to obtain second selected data;

[0060] The second fusion feature selection is specifically as follows:

[0061] S61, calculating the maximum mutual information coefficient between the feature and the label in the feature data and the average maximum information coefficient between the feature and other features;

[0062] The feature is the zth feature, where z=1, 2, ..., M 2 ; The maximum mutual information coefficient with the label is MIC3 z ;MIC3 z The value range is 0≤MIC3 z ≤1; the average maximum information coefficient with other features is MIC4 z ;MIC4 z The value range is 0≤MIC4 z ≤1;

[0063] S62, calculating the mean of the maximum mutual information coefficients of all features and labels and the mean of the average maximum information coefficients of the feature and other features;

[0064] The mean of the maximum mutual information coefficient with the label The value range is The mean of the information coefficients maximized by averaging the other features The value range is

[0065] S63, according to S62 and Use the corresponding feature selection method to select the feature data to be retained and obtain the selected features;

[0066] The S63 is specifically:

[0067] S631, using different feature selection methods to select corresponding numbers of features according to the number of feature data that you want to retain;

[0068] The number of feature data to be retained is L; the value range of L is 1≤L≤M2; the number of feature selection methods is C2, and at least one method that does not consider the correlation between features and one method that considers the correlation between features; the value range of C2 is 2≤C2; the corresponding numbers are and indivual;

[0069] S632, combining the features obtained by using different feature selection methods in S631 to obtain selected features to be supplemented;

[0070] The number of selected features to be supplemented is Ll, ​​where l represents the number of features repeatedly selected by different feature selection methods;

[0071] S633, supplement the features repeatedly selected in S632 to obtain selected features, specifically:

[0072] like Select the l non-repeating features after removing the features in S631 under the method that does not consider the correlation between features in S631 to supplement, otherwise select the l non-repeating features after removing the features in S631 under the method that considers the correlation between features in S631 to supplement;

[0073] S64, selecting the cross feature data in S53 according to the post-selection feature to obtain the data to be classified;

[0074] The number of rows of the data to be classified is X, the number of columns is L, each row corresponds to a piece of speech data, and each column corresponds to a feature;

[0075] S7, dividing the data to be classified obtained in S64 into a training set and a test set according to proportion;

[0076] S8, input the training set and labels divided in S7 into the classification learner training model, and select the model with the highest training accuracy to export;

[0077] S9, input the test set obtained in S7 into the model obtained in S8 to obtain the classification result;

[0078] So far, through S1 to S9, a classification method based on feature intersection and fusion feature selection is completed.

[0079] Beneficial Effects

[0080] Compared with the existing classification methods and devices, the classification method and device based on feature cross-selection and fusion feature selection proposed in the present invention have the following beneficial effects:

[0081] 1. The method uses two fusion feature selections, which improves the situation where the number of cross-feature dimensions is too large due to dimensional explosion when performing feature cross-feature selection;

[0082] 2. The fusion feature selection comprehensively considers the correlation between features and labels, and between features and other features, and reduces the redundancy between features while ensuring the effectiveness of the selected features;

[0083] 3. The fusion feature selection method can adaptively adjust the feature selection method used according to the above two types of correlations, and has good robustness when facing data with different characteristics;

[0084] 4. The adaptive feature crossover can complete the crossover of more than 2 types of features in one calculation process;

[0085] 5. The adaptive feature crossover outputs the crossover results of different types of features while retaining the valid features before crossover;

[0086] 6. The adaptive feature crossover can adaptively adjust the crossover method according to the linear correlation between the data to be crossed, and use nonlinear functions to calculate the features with high linear correlation, thereby enhancing the fitting degree of the nonlinear model;

[0087] 7. The nonlinear function calculation of the adaptive feature crossover is only used for one side of the crossover data, which improves the redundancy of "feature 1 crossover feature 2" and "feature 2 crossover feature 1" at the same time. The newly generated "feature 1 crossover feature 2 after nonlinear function" and "feature 2 crossover feature 1 after nonlinear function" enrich the deep features obtained by the crossover. BRIEF DESCRIPTION OF THE DRAWINGS

[0088] Figure 1 It is a structural schematic diagram of a classification device based on feature crossover and fusion feature selection of the present invention;

[0089] Figure 2 It is an overall flow chart of a classification method based on feature crossover and fusion feature selection of the present invention;

[0090] Figure 3It is a flow chart of a fusion feature selection process of a classification method based on feature intersection and fusion feature selection of the present invention;

[0091] Figure 4 It is a flow chart of an adaptive feature crossing process of a classification method based on feature crossing and fusion feature selection of the present invention;

[0092] Figure 5 It is a classification result based on human statistics expressed by a confusion matrix of a classification device based on feature crossover and fusion feature selection of the present invention;

[0093] Figure 6 The invention discloses a classification method and device based on feature crossover and fusion feature selection, which uses a confusion matrix to represent the classification result based on speech statistics. DETAILED DESCRIPTION

[0094] The classification method and device based on feature cross-fertilization and fusion feature selection proposed by the present invention are further illustrated and described in detail below in conjunction with the accompanying drawings and embodiments.

[0095] Example 1

[0096] The present invention discloses a classification device based on feature crossover and fusion feature selection, comprising a speech acquisition module, a feature extraction module, a first fusion feature selection module, an adaptive feature crossover module, a second fusion feature selection module and a model training module;

[0097] In a specific implementation, the speech acquisition module is connected to the feature extraction module, the feature extraction module is connected to the first fusion feature selection module, the first fusion feature selection module is connected to the adaptive feature cross module, the adaptive feature cross module is connected to the second fusion feature selection module, and the second fusion feature selection module is connected to the model training module;

[0098] The speech acquisition module performs digital-to-analog conversion of the speech signal and transmits the resultant speech data to the feature extraction module; the feature extraction module extracts multiple features from the speech data and transmits the obtained multiple feature data to the first fusion feature selection module; the first fusion feature selection module selects a predetermined number of features from the feature data and transmits the obtained first selected data to the adaptive feature crossover module; the adaptive feature crossover module performs adaptive feature crossover on the first selected data and transmits the obtained crossover feature data to the second fusion feature selection module; the second fusion feature selection module selects a predetermined number of features from the crossover feature data and transmits the obtained data to be classified to the model training module; the model training module performs model training using the data to be classified and finally derives the classification model with the highest training accuracy;

[0099] The adaptive feature crossover module receives the first selected data, calculates the average linear correlation coefficient between features using the linear correlation coefficient calculation unit, and determines whether the linear correlation coefficient is greater than or equal to the threshold value S. If not, that is, less than the threshold value S, the first selected data is passed as input data to the Kronecker product unit; if yes, that is, greater than or equal to the threshold value S, the first selected data is passed to the Kronecker product unit as input data after passing through the nonlinear transformation unit. In the specific implementation, S=0.5;

[0100] The feature extraction module includes a wavelet feature extraction unit, a filter feature extraction unit, and an entropy feature extraction unit; the voice acquisition module includes a microphone and an analog-to-digital converter; the microphone in the voice acquisition module is used to collect voice, and the voice collected by the microphone is converted into voice data through the analog-to-digital converter and transmitted to the feature extraction module.

[0101] The feature extraction module receives the speech data, and the feature extraction units run in parallel to extract feature data of W groups of wavelet features, F groups of filter features, and E groups of entropy features respectively;

[0102] The number of the wavelet feature extraction units is W, and W is greater than or equal to 2; in a specific implementation, W=2 or W=3;

[0103] The number of the filtering feature extraction units is F, and F is greater than or equal to 1; in a specific implementation, F=2;

[0104] The number of the entropy feature extraction units is E, and E is greater than or equal to 1; in a specific implementation, E=4;

[0105] The feature data of the W groups of wavelet features, the F groups of filter features and the E groups of entropy features are passed to the first fusion feature selection module.

[0106] The first fusion feature selection module includes a first average MIC calculation unit, a first feature selection unit and a first feature supplement unit;

[0107] The number of the first feature selection units is C1; in a specific implementation, C1=2;

[0108] The first fusion feature selection module receives the feature data, and uses the first average MIC calculation unit to calculate the average MIC of the feature and the label from the doctor's judgment, the feature and other features. According to this MIC, the C1 first feature selection units run in parallel to select the corresponding number of features of the C1 category. The first feature supplement unit supplements the features according to the difference between the number of the C1 category features after taking the union and the number of features to be retained, obtains the data after the first selection and passes it to the adaptive feature crossover module.

[0109] The adaptive feature crossover module includes a Kronecker product unit;

[0110] The Kronecker product unit calculates the Kronecker product of the input data and the data after the first selection, obtains the crossover feature data and passes it to the second feature selection module.

[0111] The second feature selection module includes a second average MIC calculation unit, a second feature selection unit and a second feature supplement unit;

[0112] The number of the second feature selection units is C2; in a specific implementation, C2=2;

[0113] The second feature selection module receives the crossover feature data, calculates the average MIC of the feature and the label, and the feature and the other features using the second average MIC calculation unit, and according to the MIC, the C2 second feature selection units run in parallel to select the corresponding number of features of the C2 class, and the second feature supplementation unit supplements the features according to the difference between the number of the C2 class features after the union and the number of features to be retained, obtains the data to be classified and passes it to the model training module;

[0114] The model training module includes several classification models;

[0115] The model training module receives the data to be classified and puts it into several classification models for training, and uses the model with the highest training accuracy as the output of the model training module.

[0116] Example 2

[0117] As a second aspect of the present invention, a classification method based on feature crossover and fusion feature selection is proposed, comprising the following steps:

[0118] S1, performing analog-to-digital conversion on the collected voice to obtain voice data and adding a label to each voice data;

[0119] S2, extracting several types of features from the speech data;

[0120] S3, concatenating the feature data corresponding to each voice data, obtaining a concatenated vector and then expressing it as a concatenated two-dimensional matrix;

[0121] S4, according to the label, perform the first fusion feature selection on all feature data in the spliced ​​two-dimensional matrix to obtain the first selected data, specifically: according to the mean of the maximum mutual information coefficients of all features and labels and the mean of the average maximum mutual information coefficients of the feature and other features, select the number of feature data you want to retain and use different feature selection methods to select the corresponding number of features respectively; then use the features obtained by different feature selection methods to take the union set to obtain the selected features to be supplemented; supplement the repeatedly selected features to obtain the selected features; then use different feature selection methods to select feature data to retain and obtain the selected features; then select the corresponding feature data according to the selected features to obtain the first selected data, including the following sub-steps:

[0122] S41, calculating the maximum mutual information coefficient between the feature and the label in the feature data and the average maximum information coefficient between the feature and other features;

[0123] S42, calculating the mean of the maximum mutual information coefficients of all features and labels and the mean of the average maximum information coefficients of the feature and other features;

[0124] S43, selecting feature data to be retained using a corresponding feature selection method according to the result obtained in S42, to obtain selected features;

[0125] S44, selecting the feature data in S3 according to the selected feature to obtain the first selected data;

[0126] S5, performing adaptive feature crossover on the data after the first selection to obtain crossover feature data;

[0127] The adaptive feature crossover is: if the linear correlation coefficient between features is greater than a threshold, nonlinear transformation is performed first and then feature crossover is performed; otherwise, feature crossover is performed directly;

[0128] S6, performing a second fusion feature selection on the cross features of the cross feature data according to the labels to obtain data to be classified;

[0129] S7, dividing the data to be classified into a training set and a test set in proportion;

[0130] S8, input the training set and labels into the classification learner training model, select and export the model with the highest training accuracy;

[0131] S9. Input the test set into the model obtained in S8 to obtain the classification result.

[0132] The first fusion feature selection in S4 or the second fusion feature selection in S6 is specifically:

[0133] SA1, calculate the maximum mutual information coefficient between the feature and the label in the feature data and the average maximized information coefficient between the feature and other features;

[0134] SA2, calculate the mean of the maximum mutual information coefficients of all features and labels and the mean of the average maximized information coefficients of the feature and other features;

[0135] SA3, according to the results of SA2, use the corresponding feature selection method to select the feature data to be retained, and obtain the selected features;

[0136] SA4. Select cross-feature data based on the selected features to obtain data to be classified.

[0137] S1: The number of voice data is X; the value range of X is 2≤X;

[0138] The number of feature types in S2 is N; the value range of N is 1≤N;

[0139] The nth feature of each voice data has Y n feature data and the number of types of the feature data is the same as the number of feature types; wherein n = 1, 2, ..., N; the total number of the features of the several types is

[0140] The length of the concatenated vector S3 is The spliced ​​two-dimensional matrix is ​​a two-dimensional matrix with X rows and Y columns, including feature data of all X voice data, each row corresponds to a voice data, and each column corresponds to a feature; if it is the first fusion feature selection, the SA includes the following sub-steps:

[0141] SA1 The feature data includes Y features, denoted as y=1, 2, ..., Y; the maximum mutual information coefficient with the label is MIC1 y ;MIC1 y The value range is 0≤MIC1 y ≤1; the average maximum information coefficient with other features is MIC2 y ;MIC2 y The value range is 0≤MIC2 y ≤1;

[0142] The mean of the maximum mutual information coefficients with the labels described in SA2 The value range is The mean of the information coefficients maximized by averaging the other features The value range is

[0143] The SA3 is specifically as follows: SA31, using different feature selection methods to select corresponding numbers of features according to the number of feature data that are desired to be retained;

[0144] The number of feature data to be retained is M; the value range of M is 1≤M≤Y; the number of feature selection methods is C1, and at least one method that does not consider the correlation between features and one method that considers the correlation between features; the value range of C1 is 2≤C1; the corresponding numbers are and indivual;

[0145] SA32, taking the union of the features obtained by using different feature selection methods in SA31 to obtain the selected features to be supplemented;

[0146] The number of selected features to be supplemented is Mm, where m represents the number of features repeatedly selected by different feature selection methods;

[0147] SA33, supplement the features repeatedly selected in SA32 to obtain the selected features, specifically: Select m non-repeating features after removing the features of S431 in the method without considering the correlation between features in SA31 for supplementation; otherwise select m non-repeating features after removing the features in S431 in the method with considering the correlation between features in SA31 for supplementation;

[0148] The number of rows of data after the first selection described in SA4 is X, and the number of columns is M. Each row corresponds to a piece of speech data, and each column corresponds to a feature.

[0149] S5 is specifically as follows: S51, calculating the linear correlation coefficient between every two columns of feature data of the first selected data and then calculating the mean of all linear correlation coefficients;

[0150] S51 linear correlation coefficient r ij ,i,j=1,2,...,M,j≠i;r ij The value range is 0≤r ij ≤1; the mean of the linear correlation coefficient The value range is

[0151] S52, if S51 If the value is less than the threshold value S, the Kronecker product of each row of data in S34 with itself is calculated respectively; otherwise, the result of each row of data passing through the nonlinear function is calculated first, and then the Kronecker product of the result and the row of data is calculated;

[0152] The threshold S has a value range of 0≤S≤1; the Kronecker product is a square matrix with a dimension of M in both cases;

[0153] S53, arranging the square matrix elements corresponding to each row of data obtained in S52 into row vectors, so that the characteristic data corresponding to all physiological data can be represented by a two-dimensional matrix, which is called cross characteristic data;

[0154] The length of the row vector is M 2 The number of rows of the cross-feature data is X, and the number of columns is M 2 , each row corresponds to a piece of speech data, and each column corresponds to a cross feature.

[0155] If it is the second fusion feature selection, the sub-steps of S6 are S61, S62, S63 and S64 in order, specifically:

[0156] S61 The feature is the zth feature, where z=1, 2, ..., M 2 ; The maximum mutual information coefficient with the label is MIC3 z ;MIC3 z The value range is 0≤MIC3 z ≤1; the average maximum information coefficient with other features is MIC4 z ;MIC4 z The value range is 0≤MIC4 z ≤1;

[0157] S62: Mean value of the maximum mutual information coefficient with the label The value range is The mean of the information coefficients maximized by averaging the other features The value range is

[0158] The S63 is specifically:

[0159] S631, using different feature selection methods to select corresponding numbers of features according to the number of feature data that you want to retain;

[0160] The number of feature data to be retained is L; the value range of L is 1≤L≤M2; the number of feature selection methods is C2, and at least one method that does not consider the correlation between features and one method that considers the correlation between features; the value range of C2 is 2≤C2; the corresponding numbers are and indivual;

[0161] S632, combining the features obtained by using different feature selection methods in S631 to obtain selected features to be supplemented;

[0162] The number of selected features to be supplemented is Ll, ​​where l represents the number of features repeatedly selected by different feature selection methods;

[0163] S633, supplement the features repeatedly selected in S632 to obtain selected features, specifically:

[0164] like Select the l non-repeating features after removing the features in S631 under the method that does not consider the correlation between features in S631 to supplement, otherwise select the l non-repeating features after removing the features in S631 under the method that considers the correlation between features in S631 to supplement;

[0165] S64: The number of rows of the data to be classified is X, and the number of columns is L. Each row corresponds to a piece of speech data, and each column corresponds to a feature.

[0166] When implementing it specifically, Figure 1 The present invention is a structural diagram of a classification device based on feature crossover and fusion feature selection. The present invention is a classification device based on feature crossover and fusion feature selection, comprising a speech acquisition module, a feature extraction module, a first fusion feature selection module, an adaptive feature crossover module, a second fusion feature selection module and a model training module.

[0167] The speech collection module is connected to the feature extraction module, the feature extraction module is connected to the first fusion feature selection module, the first fusion feature selection module is connected to the adaptive feature cross module, the adaptive feature cross module is connected to the second fusion feature selection module, and the second fusion feature selection module is connected to the model training module.

[0168] The speech acquisition module performs digital-to-analog conversion of the speech signal and transmits the resultant speech data to the feature extraction module; the feature extraction module extracts multiple features from the speech data and transmits the obtained multiple feature data to the first fusion feature selection module; the first fusion feature selection module selects a predetermined number of features from the feature data and transmits the obtained first selected data to the adaptive feature crossover module; the adaptive feature crossover module performs adaptive feature crossover on the first selected data and transmits the obtained crossover feature data to the second fusion feature selection module; the second fusion feature selection module selects a predetermined number of features from the crossover feature data and transmits the obtained data to be classified to the model training module; the model training module performs model training using the above data to be classified and finally derives the classification model with the highest training accuracy;

[0169] The adaptive feature crossover module receives the data after the first selection, calculates the average linear correlation coefficient between features using the linear correlation coefficient calculation unit, and determines whether the linear correlation coefficient is greater than or equal to a threshold value S. If not, that is, less than the threshold value S, the data after the first selection is passed as input data to the Kronecker product unit; if yes, that is, greater than or equal to the threshold value S, the data after the first selection is passed through the nonlinear transformation unit and then passed as input data to the Kronecker product unit; the S is greater than or equal to 1 / 3 and less than or equal to 1. Specifically in this embodiment, S is 1 / 3, and can also be 0.5.

[0170] The feature extraction module includes a wavelet feature extraction unit, a filter feature extraction unit, and an entropy feature extraction unit; the voice acquisition module includes a microphone and an analog-to-digital converter; the microphone in the voice acquisition module is used to collect voice, and the voice collected by the microphone is converted into voice data through the analog-to-digital converter and transmitted to the feature extraction module.

[0171] Furthermore, the feature extraction module receives the speech data, and the three feature extraction units are operated in parallel to extract feature data of W groups of wavelet features, F groups of filter features and E groups of entropy features respectively;

[0172] The number of the wavelet feature extraction units is W, and W is greater than or equal to 2. Specifically in this embodiment, W is 2 or 5.

[0173] The number of the filtering feature extraction units is F, and F is greater than or equal to 1; specifically in this embodiment, F is 1 or 2.

[0174] The number of the entropy feature extraction units is E, and E is greater than or equal to 1; specifically in this embodiment, E is 1, and may also be 3;

[0175] The feature data of the W groups of wavelet features, the F groups of filter features and the E groups of entropy features are passed to the first fusion feature selection module.

[0176] The first fusion feature selection module includes a first average MIC calculation unit, a first feature selection unit and a first feature supplement unit;

[0177] Preferably, the number of the first feature selection units is C1, and C1 is greater than or equal to 2; specifically in this embodiment, C1 is 2, and may also be 4;

[0178] The first fusion feature selection module receives the above feature data, and uses the first average MIC calculation unit to calculate the average MIC of the feature and the label from the doctor's judgment, the feature and other features. According to this MIC, the C1 first feature selection units run in parallel to select the corresponding number of features of the C1 category. The first feature supplement unit supplements the features according to the difference between the number of the above C1 category features after taking the union and the number of features to be retained, obtains the data after the first selection and passes it to the adaptive feature crossover module.

[0179] The adaptive feature crossover module includes a linear correlation coefficient calculation unit, a nonlinear transformation unit, and a Kronecker product unit;

[0180] The Kronecker product unit calculates the Kronecker product of the input data and the data after the first selection, obtains the crossover feature data and passes it to the second feature selection module;

[0181] The value range of the threshold S is 1 / 3≤S≤1.

[0182] The second feature selection module includes a second average MIC calculation unit, a second feature selection unit and a second feature supplement unit;

[0183] The number of the second feature selection units is C2, and C2 is greater than or equal to 2; specifically in this embodiment, C2 is 2, and may also be 5;

[0184] The second feature selection module receives the crossover feature data, calculates the average MIC of the feature and the label, and the feature and other features using the second average MIC calculation unit, and according to the MIC, the C2 second feature selection units run in parallel to select the corresponding number of features of the C2 category, and the second feature supplementation unit supplements the features according to the difference between the number of the C2 category features after the union and the number of features to be retained, obtains the data to be classified and passes it to the model training module;

[0185] The model training module includes several classification models;

[0186] The model training module receives the data to be classified and puts it into several classification models for training, and takes the model with the highest training accuracy as the output of the model training module.

[0187] Figure 2 It is an overall flow chart of a classification method based on feature crossover and fusion feature selection of the present invention; the method specifically includes:

[0188] S1. Perform analog-to-digital conversion on the collected voice to obtain voice data and add a label to each voice data according to the doctor's diagnosis result.

[0189] The number of voice data is 546, collected from 78 subjects, including 24 healthy people and 54 patients with depression, and 7 voice data are collected from each subject.

[0190] S2. Extract several types of features from the speech data.

[0191] The number of feature types is 4, which are wavelet features based on bior2.4, wavelet features based on bior3.7, features based on RASTA filtering, and entropy features;

[0192] Each piece of speech data has 150 feature data based on the wavelet feature of bior2.4; each piece of speech data has 150 feature data based on the wavelet feature of bior3.7; each piece of speech data has 615 feature data based on the feature of RASTA filtering; each piece of speech data has 13 feature data based on the entropy feature; the total number of the several types of features is 928.

[0193] S3. Concatenate the feature data corresponding to each voice data, obtain the concatenated vector and then express it as a concatenated two-dimensional matrix.

[0194] The length of the concatenated vector is 928; the concatenated two-dimensional matrix is ​​a two-dimensional matrix with 546 rows and 928 columns, including feature data of all 546 voice data; each row corresponds to a voice data, and each column corresponds to a feature;

[0195] S4. Perform the first fusion feature selection on all features in the spliced ​​two-dimensional matrix according to the labels to obtain feature data after the first selection.

[0196] Figure 3 It is a flow chart of an adaptive feature crossing process of a classification method based on feature crossing and fusion feature selection of the present invention.

[0197] The first fusion feature selection is specifically as follows:

[0198] S41, calculating the maximum mutual information coefficient between the feature and the label in the feature data and the average maximum information coefficient between the feature and other features;

[0199] The feature is the yth feature, where y=1, 2, ..., 928; the maximum mutual information coefficient with the label is MIC1 y ;MIC1 y The value range is 0≤MIC1 y ≤1; the average maximum information coefficient with other features is MIC2 y ;MIC2 y The value range is 0≤MIC2 y ≤1.

[0200] S42. Calculate the mean of the maximum mutual information coefficients of all features and labels and the mean of the average maximized information coefficients of the feature and other features.

[0201] The mean of the maximum mutual information coefficient with the label The value range is The mean of the information coefficients maximized by averaging the other features The value range is

[0202] S43, according to S42 and Use the corresponding feature selection method to select the feature data to be retained and obtain the selected features;

[0203] The S43 is specifically:

[0204] S431, using different feature selection methods to select corresponding numbers of features according to the number of feature data that you want to retain;

[0205] The number of feature data to be retained is 120; the number of feature selection methods is 2, which are a feature selection method based on neighbor component analysis (NCA) and a feature selection method based on minimum redundancy maximum correlation (mRMR); the corresponding numbers are and indivual;

[0206] S432, combining the features obtained by using different feature selection methods in S431 to obtain selected features to be supplemented;

[0207] The number of selected features to be supplemented is 120-m, where m represents the number of features repeatedly selected by different feature selection methods;

[0208] S433, supplement the features repeatedly selected in S432 to obtain selected features, specifically:

[0209] like Select m non-repeating features after removing the features of S431 under the feature selection method based on NCA for supplementation; otherwise select m non-repeating features after removing the features in S431 under the feature selection method based on mRMR for supplementation;

[0210] S44, selecting the feature data in S3 according to the selected feature to obtain the first selected data;

[0211] After the first selection, the data has 546 rows and 120 columns, each row corresponds to a piece of speech data, and each column corresponds to a feature.

[0212] Figure 4 It is a flow chart of a cross-feature data selection process of a classification method based on feature cross-feature and fusion feature selection of the present invention.

[0213] S5, performing adaptive feature crossover on the first selected data in S44 to obtain crossover feature data, specifically:

[0214] S51, calculating the linear correlation coefficient between every two columns of feature data of the first selected data in S44 and then calculating the mean of all linear correlation coefficients;

[0215] The linear correlation coefficient r ij ,i,j=1,2,...,120,j≠i;r ij The value range is 0≤r ij ≤1; the mean of the linear correlation coefficient The value range is

[0216] S52, if S51 If the value is less than the threshold value S, the Kronecker product of each row of data in S34 with itself is calculated respectively; otherwise, the result of each row of data passing through the nonlinear function is calculated first, and then the Kronecker product of the result and the row of data is calculated;

[0217] The threshold value S is 0.5; the Kronecker product is a square matrix with a dimension of 120 in both cases;

[0218] S53, arranging the square matrix elements corresponding to each row of data obtained in S52 into row vectors, so that the characteristic data corresponding to all physiological data can be represented by a two-dimensional matrix, which is called cross characteristic data;

[0219] The row vector length is 120 2 The number of rows of the cross-feature data is 546, and the number of columns is 120 2 , each row corresponds to a voice data, and each column corresponds to a cross feature;

[0220] S6, performing a second fusion feature selection on the cross features of the cross feature data obtained in S53 according to the labels to obtain second selected data;

[0221] The second fusion feature selection is specifically as follows:

[0222] S61, calculating the maximum mutual information coefficient between the feature and the label in the feature data and the average maximum information coefficient between the feature and other features;

[0223] The feature is the zth feature, where z=1, 2, ..., 1202 ; The maximum mutual information coefficient with the label is MIC3 z ;MIC3 z The value range is 0≤MIC3 z ≤1; the average maximum information coefficient with other features is MIC4 z ;MIC4 z The value range is 0≤MIC4 z ≤1;

[0224] S62, calculating the mean of the maximum mutual information coefficients between the feature and the label in all feature data and the mean of the average maximized information coefficients between the feature and other features;

[0225] The mean of the maximum mutual information coefficient with the label The value range is The mean of the information coefficients maximized by averaging the other features The value range is

[0226] S63, according to S62 and Use the corresponding feature selection method to select the feature data to be retained and obtain the selected features;

[0227] The S63 is specifically:

[0228] S631, using different feature selection methods to select corresponding numbers of features according to the number of feature data that you want to retain;

[0229] The number of feature data to be retained is 120; the number of feature selection methods is 2, which are a feature selection method based on neighbor component analysis (NCA) and a feature selection method based on minimum redundancy maximum correlation (mRMR); the corresponding numbers are and indivual;

[0230] S632, combining the features obtained by using different feature selection methods in S631 to obtain selected features to be supplemented;

[0231] The number of selected features to be supplemented is 120-1, where 1 represents the number of features repeatedly selected by different feature selection methods;

[0232] S633, supplement the features repeatedly selected in S632 to obtain selected features, specifically:

[0233] like Select l non-repeating features after removing the features of S631 under the feature selection method based on NCA for supplementation, otherwise select l non-repeating features after removing the features in S631 under the feature selection method based on mRMR for supplementation;

[0234] S64, selecting the cross feature data in S53 according to the post-selection feature to obtain the data to be classified;

[0235] The number of rows of the data to be classified is 546, the number of columns is 120, each row corresponds to a piece of speech data, and each column corresponds to a feature;

[0236] S7, dividing the data to be classified obtained in S64 into a training set and a test set in a ratio of 8:2;

[0237] S8, input the training set and labels divided in S7 into the classification learner training model, and select the model with the highest training accuracy to export;

[0238] S9. Input the test set obtained in S7 into the model obtained in S8 to obtain the classification result of each speech; count the classification results of the 9 speech of each subject, if at least 5 are classified as depression, it is considered that the subject suffers from depression.

[0239] The classification results are compared with the labels to generate a confusion matrix and the accuracy is calculated as the ratio of the number of people or voice items whose classification results are the same as the labels to the total number of people or voice items in the test set. The confusion matrix based on person statistics and voice statistics is shown in the figure. Figure 5 It is a classification result based on human statistics expressed by a confusion matrix of a classification device based on feature crossover and fusion feature selection of the present invention; Figure 5 0 is mdd; 1 is hc; among mdd, 11 people made correct judgments; among hc, 5 people also made correct judgments; it can be obtained that the accuracy rate based on human statistics is 100%. Figure 6 The classification method and device based on feature crossover and fusion feature selection of the present invention shows the classification result based on speech statistics by using confusion matrix, and the accuracy rate based on speech statistics is 96.43%. Figure 6 0 is mdd; 1 is hc; in mdd, 73 speech sounds were judged correctly; 4 speech sounds were judged incorrectly; in hc, 36 speech sounds were judged correctly.

[0240] The above is only a preferred embodiment of the present invention, and the present invention should not be limited to the contents disclosed in the embodiment and the drawings. Any equivalent or modification completed without departing from the spirit disclosed in the present invention shall fall within the scope of protection of the present invention.

Claims

1. A classification device based on feature crossover and fusion feature selection, characterized in that: It includes a speech acquisition module, a feature extraction module, a first fusion feature selection module, an adaptive feature cross module, a second fusion feature selection module and a model training module; The voice acquisition module converts the voice signal into voice data by digital-to-analog conversion and transmits it to the feature extraction module; The feature extraction module extracts multiple features from the speech data, and passes the obtained multiple feature data to the first fusion feature selection module; the first fusion feature selection module selects a predetermined number of features from the feature data, and passes the obtained first selected data to the adaptive feature crossover module; the adaptive feature crossover module performs adaptive feature crossover on the first selected data, and passes the obtained crossover feature data to the second fusion feature selection module; the second fusion feature selection module selects a predetermined number of features from the crossover feature data, and passes the obtained data to be classified to the model training module; the model training module performs model training using the data to be classified, and finally derives a classification model with the highest training accuracy; The adaptive feature crossover module receives the data after the first selection, calculates the average linear correlation coefficient between features using a linear correlation coefficient calculation unit, and determines whether the linear correlation coefficient is greater than or equal to a threshold value S. If not, that is, less than the threshold value S, the data after the first selection is passed as input data to the Kronecker product unit; if so, that is, greater than or equal to the threshold value S, the data after the first selection is passed through a nonlinear transformation unit and then passed as input data to the Kronecker product unit.

2. A classification device based on feature crossover and fusion feature selection according to claim 1, characterized in that: The feature extraction module includes a wavelet feature extraction unit, a filter feature extraction unit, and an entropy feature extraction unit; the voice collection module includes a microphone and an analog-to-digital converter; the microphone in the voice collection module is used to collect voice, and the voice collected by the microphone is converted into voice data through the analog-to-digital converter and passed to the feature extraction module.

3. The classification device based on feature crossover and fusion feature selection according to claim 1, characterized in that: The feature extraction module receives the speech data, and the feature extraction units run in parallel to extract feature data of W groups of wavelet features, F groups of filter features, and E groups of entropy features respectively; The number of the wavelet feature extraction units is W, and W is greater than or equal to 2; The number of the filtering feature extraction units is F, and F is greater than or equal to 1; The number of the entropy feature extraction units is E, and E is greater than or equal to 1; The feature data of the W groups of wavelet features, the F groups of filter features and the E groups of entropy features are passed to the first fusion feature selection module.

4. The classification device based on feature crossover and fusion feature selection according to claim 1, characterized in that: The first fusion feature selection module includes a first average MIC calculation unit, a first feature selection unit and a first feature supplement unit; The number of the first feature selection units is C1; The first fusion feature selection module receives the feature data, and uses the first average MIC calculation unit to calculate the average MIC of the feature and the label from the doctor's judgment, the feature and other features. According to this MIC, the C1 first feature selection units run in parallel to select the corresponding number of features of the C1 category. The first feature supplement unit supplements the features according to the difference between the number of the C1 category features after taking the union and the number of features to be retained, obtains the data after the first selection and passes it to the adaptive feature crossover module.

5. The classification device based on feature crossover and fusion feature selection according to claim 1, characterized in that: The adaptive feature crossover module includes a Kronecker product unit; The Kronecker product unit calculates the Kronecker product of the input data and the data after the first selection, obtains the crossover feature data and passes it to the second feature selection module.

6. The classification device based on feature crossover and fusion feature selection according to claim 1, characterized in that: The second feature selection module includes a second average MIC calculation unit, a second feature selection unit and a second feature supplement unit; The number of the second feature selection units is C2; The second feature selection module receives the crossover feature data, calculates the average MIC of the feature and the label, and the feature and the other features using the second average MIC calculation unit, and according to the MIC, the C2 second feature selection units run in parallel to select the corresponding number of features of the C2 class, and the second feature supplementation unit supplements the features according to the difference between the number of the C2 class features after the union and the number of features to be retained, obtains the data to be classified and passes it to the model training module; The model training module includes several classification models; The model training module receives the data to be classified and puts it into several classification models for training, and uses the model with the highest training accuracy as the output of the model training module.

7. A classification method based on feature crossover and fusion feature selection, characterized in that: include: S1, performing analog-to-digital conversion on the collected voice to obtain voice data and adding a label to each voice data; S2, extracting several types of features from the speech data; S3, concatenating the feature data corresponding to each voice data, obtaining a concatenated vector and then expressing it as a concatenated two-dimensional matrix; S4, performing a first fusion feature selection on all feature data in the spliced ​​two-dimensional matrix according to the label to obtain data after the first selection; S5, performing adaptive feature crossover on the data after the first selection to obtain crossover feature data; The adaptive feature crossover is: if the linear correlation coefficient between features is greater than a threshold, nonlinear transformation is performed first and then feature crossover is performed; otherwise, feature crossover is performed directly; S6, performing a second fusion feature selection on the cross features of the cross feature data according to the labels to obtain data to be classified; S7, dividing the data to be classified into a training set and a test set in proportion; S8, input the training set and labels into the classification learner training model, select and export the model with the highest training accuracy; S9. Input the test set into the model obtained in S8 to obtain the classification result.

8. The classification method based on feature crossover and fusion feature selection according to claim 7, characterized in that: The first fusion feature selection or the second fusion feature selection is specifically: SA1, calculate the maximum mutual information coefficient between the feature and the label in the feature data and the average maximized information coefficient between the feature and other features; SA2, calculate the mean of the maximum mutual information coefficients of all features and labels and the mean of the average maximized information coefficients of the feature and other features; SA3, according to the results of SA2, use the corresponding feature selection method to select the feature data to be retained, and obtain the selected features; SA4. Select cross-feature data based on the selected features to obtain data to be classified.

9. A classification method based on feature crossover and fusion feature selection according to claim 8, characterized in that: S1: The number of voice data is X; the value range of X is 2≤X; The number of feature types in S2 is N; the value range of N is 1≤N; The nth feature of each voice data has Y n feature data and the number of types of the feature data is the same as the number of feature types; wherein n = 1, 2, ..., N; the total number of the features of the several types is The length of the concatenated vector S3 is The spliced ​​two-dimensional matrix is ​​a two-dimensional matrix with X rows and Y columns, including feature data of all X voice data, each row corresponds to a voice data, and each column corresponds to a feature; SA1 The feature data includes Y features, denoted as y=1, 2, ..., Y; the maximum mutual information coefficient with the label is MIC1 y ;MIC1 y The value range is 0≤MIC1 y ≤1; the average maximum information coefficient with other features is MIC2 y ;MIC2 y The value range is 0≤MIC2 y ≤1; The mean of the maximum mutual information coefficients with the labels described in SA2 The value range is The mean of the information coefficients maximized by averaging the other features The value range is The SA3 is specifically: SA31. Use different feature selection methods to select the corresponding number of features according to the number of feature data you want to retain; The number of feature data that is desired to be retained is M; the value range of M is 1≤M≤Y; The number of feature selection methods is C1, and at least one method that does not consider the correlation between features and one method that considers the correlation between features; the value range of C1 is 2≤C1; the corresponding numbers are and indivual; SA32, taking the union of the features obtained by using different feature selection methods in SA31 to obtain the selected features to be supplemented; The number of selected features to be supplemented is Mm, where m represents the number of features repeatedly selected by different feature selection methods; SA33, supplement the features repeatedly selected in SA32 to obtain the selected features, specifically: Select m non-repeating features after removing the features of S431 in the method without considering the correlation between features in SA31 for supplementation; otherwise select m non-repeating features after removing the features in S431 in the method with considering the correlation between features in SA31 for supplementation; The number of rows of data after the first selection described in SA4 is X, and the number of columns is M. Each row corresponds to a piece of speech data, and each column corresponds to a feature.

10. A classification method based on feature crossover and fusion feature selection according to claim 7, characterized in that: S5 is specifically as follows: S51, calculating the linear correlation coefficient between every two columns of feature data of the first selected data and then calculating the mean of all linear correlation coefficients; S51 linear correlation coefficient r ij ,i,j=1,2,...,M,j≠i;r ij The value range is 0≤r ij ≤1; the mean of the linear correlation coefficient The value range of is 0≤r≤1; S52, if S51 If the value is less than the threshold value S, the Kronecker product of each row of data in S34 with itself is calculated respectively; otherwise, the result of each row of data passing through the nonlinear function is calculated first, and then the Kronecker product of the result and the row of data is calculated; The threshold S has a value range of 0≤S≤1; the Kronecker product is a square matrix with a dimension of M in both cases; S53, arranging the square matrix elements corresponding to each row of data obtained in S52 into row vectors, so that the characteristic data corresponding to all physiological data can be represented by a two-dimensional matrix, which is called cross characteristic data; The length of the row vector is M 2 The number of rows of the cross-feature data is X, and the number of columns is M 2 , each row corresponds to a piece of speech data, and each column corresponds to a cross feature.

Images