CN106156789A - Towards the validity feature sample identification techniques strengthening grader popularization performance - Google Patents

Abstract

The invention discloses an effective feature sample identification technology for enhancing classifier generalization performance, which is characterized in that the method comprises the following steps: 1) establishment of classifier generalization performance evaluation index; 2) construction of fuzzy clustering criterion; 3 ) primary clustering division of feature sample set; 4) definition of intra-class average distance and inter-class average distance; 5) establishment of initial cluster optimization criteria; 6) secondary cluster identification of feature sample set. The invention has the beneficial effects that the method is reasonable in design, simple in use, effectively removes noise points or wild points, and has a high recognition rate of feature samples.

Description

Efficient Feature Sample Recognition Techniques for Enhancing Classifier Generalization Performance

technical field

Based on the signal processing theory, the present invention proposes an effective feature sample identification method on the basis of data cluster analysis, uses the cluster analysis mode to automatically divide characteristics, eliminates wild points or noise points in feature data, and achieves the purpose of feature data purification. On this basis, the generalization performance of the support vector machine classifier is improved. This calculation method lays the foundation for solving the precise pattern recognition and classification problems involved in the field of mechanical fault diagnosis.

Background technique

The support vector machine (SVM) method based on statistical learning theory has great advantages in pattern classification and has been successfully applied to fault diagnosis. Theoretically, the optimal classification surface of SVM is determined by the support vectors located at the edge of the class, and the wild (value) points or noise (sound) points near the edge of the class are often mixed with valid samples, resulting in the obtained classification surface It is not optimal, which affects the generalization performance of the classifier ^[1,2] .

In actual fault diagnosis applications, external interference around the object to be diagnosed in signal detection and internal noise of the acquisition system may introduce noise interference into the original observation data; abnormal or faulty sensor devices, abnormal fluctuations in force or motion in the system Or just changes in operating conditions may also produce abnormal observation outliers. If these noises or outliers in the original data are not properly processed, they will enter the feature space along with the feature extraction, forming noise or outliers that obviously deviate from the overall class features. In addition, there are many negative factors that affect fault diagnosis, such as the redundancy of sensing observation information caused by the scattering and reverberation effects of vibration transmitted in the mechanical structure, and the excessively high feature dimension selected in the feature extraction process. Information redundancy will cause difficulties in subsequent feature extraction and further amplify the negative effects of noise or outliers; if the feature dimension is too high, it will make it more difficult to estimate the statistical characteristics of the sample, thereby reducing the generalization ability or generalization of the classifier Capability ^[3] . Therefore, the characteristic data must first be purified to achieve the purpose of effective diagnosis.

Contents of the invention

The purpose of the present invention is to solve the above problems and develop an effective feature sample identification technology for enhancing the generalization performance of classifiers.

Achieving the above object The technical solution of the present invention is an effective feature sample identification technology for enhancing the generalization performance of classifiers, characterized in that the method includes the following steps:

1) Establishment of classifier generalization performance evaluation index;

2) Construction of fuzzy clustering criteria;

3) One-time clustering and division of feature sample set;

4) The definition of the average distance within a class and the average distance between classes;

5) Establishment of initial clustering optimization criteria;

6) Secondary clustering identification of feature sample set.

The establishment calculation formula of described classifier generalization performance evaluation index is:

R(w)＝ _Remp (w)+Φ(h/l),

h≤min([r ² a ² ],n)+1.

where Φ(·) is the confidence risk function, h is the VC dimension of the classification function, and l is the number of training samples. It can be seen that the real risk R(w) is composed of two parts: the empirical risk _Remp (w) and the confidence risk Φ(·). [·] indicates the integer part. r contains all high-dimensional space mapping points The minimum hypersphere radius of .

The formula for constructing the fuzzy clustering criterion is:

In the formula, d _ik ＝||x _k -v _i| | is the distance between the sample x _k and the cluster center v _i , which is generally measured by Euclidean distance. m is the fuzzy weighting index, usually m=2. J _FCM (U, V) represents the sum of squares of weighted distances from various samples to the cluster center, and the weight is the mth power of the membership degree of sample x _k to the i-th class.

The calculation formula of the primary clustering division of the feature sample set is:

In the formula, the number of clusters c, the fuzzy weight index m and the initial membership degree matrix U ⁰ are set, and the number of iteration steps l=0. For a given stop value ε>0, iterative calculation until max{|u _ik ^l -u _ik ^l-1 |}<ε, the algorithm terminates; otherwise l=l+1, the algorithm continues to execute.

The definition calculation formula of the average distance within the class and the average distance between classes is:

In the formula is the number of pairwise combinations of data samples in the cluster {X _O }. v _i and v _j are the centers of the i-th cluster {X _i } and the j-th cluster {X _j } respectively. The number of combinations that number pairwise combinations for c clusters.

The formula for establishing the initial clustering optimization criterion is:

In the formula, {X _f } is an initial effective cluster with a capacity of n _f and a center of v _f which is composed of effective feature samples contained in c clusters (usually c≥3), and {X _n } is mainly composed of The initial invalid cluster {X _n } with capacity n _n and center v _n is composed of noise or wild points.

The formula for calculating the secondary clustering identification of the feature sample set is:

In the formula, {X _s } is a combined subset with capacity n _s and center v _s extracted from {X _d }, which satisfies the minimization criteria. After effective sample extraction, the remaining data samples in {X _d } form a subset {X _t }, which is merged into the invalid cluster {X _n } composed of noise (wild) points. Equation (18) implements the secondary division process of invalid clusters, where _xi is the data sample in the invalid cluster {X _n } formed after the main division. X _near is the data sample closest to the center v _n of the invalid cluster {X _n } in the valid cluster {X _f }.

Description of drawings

Fig. 1 is a schematic flow chart of the effective feature sample identification technology for enhancing classifier generalization performance according to the present invention

Figure 2 is a schematic diagram of SVM classifier generalization performance evaluation

Figure 3 Hypersphere description in feature space

Fig.4 Recognition results of effective feature samples in normal state

Figure 5. Identification results of effective feature samples of tooth damage

Figure 6 The recognition results of the effective feature samples of machine base looseness

detailed description

The present invention is described in detail below in conjunction with the accompanying drawings. Fig. 1 is a schematic flow chart of the effective feature sample recognition technology for enhancing the generalization performance of the classifier according to the present invention. The feature samples in the space are pre-cleaned.

This technical solution uses the effective feature sample identification of the three types of gearbox normal state, gear tooth damage and machine base looseness as an example to illustrate the feature sample purification and pretreatment process. The basic principle is: Structural Risk Minimization (SRM) based on Statistical Learning Theory ) principle to maximize the generalization performance of the SVM classifier, perform two purification processes on multiple failure mode feature samples according to the principle of hierarchical processing, and obtain effective feature samples for classifier design. The generalization performance evaluation principle of the SVM classifier is as follows: figure 2. which is

R(w)＝ _Remp (w)+Φ(h/l),

h≤min([r ² a ² ],n)+1.

where Φ(·) is the confidence risk function, h is the VC dimension of the classification function, and l is the number of training samples. It can be seen that the real risk R(w) is composed of two parts: the empirical risk _Remp (w) and the confidence risk Φ(·). [·] indicates the integer part. r contains all high-dimensional space mapping points The minimum hypersphere radius of . The description of the hypersphere in the feature space is shown in Figure 3.

Example 1

Recognition of Valid Feature Samples in Normal State

The clustering criteria are established sequentially for the feature sample set in normal state, and two consecutive clustering divisions are carried out. The identification results of effective feature samples are shown in Figure 4.

The establishment formula of the clustering criterion is:

The primary clustering formula of the normal state feature sample set is:

The definition of the average distance within a class and the average distance between classes in a normal state is:

The establishment formula of the initial clustering optimization criterion of normal state feature samples is:

The calculation formula for the secondary clustering identification of the normal state feature sample set is:

Example 2

Identification of Effective Feature Samples for Gear Tooth Destruction

The clustering criterion is established sequentially for the tooth damage feature sample set, and two consecutive clustering divisions are carried out. The effective feature sample identification results are shown in Fig. 5.

The establishment formula of the clustering criterion is:

The primary clustering formula of the gear tooth damage feature sample set is:

The definition formulas of the average distance within a class and the average distance between classes of tooth damage are:

The establishment formula of the initial clustering optimization criterion for gear tooth damage feature samples is:

The formula for secondary clustering identification of tooth damage feature sample set is:

Example 3

Recognition of Effective Feature Samples for Machine Base Looseness

The clustering criterion is established sequentially for the loose feature sample set of the machine base, and two consecutive clustering divisions are carried out. The effective feature sample identification results are shown in Figure 6.

The establishment formula of the clustering criterion is:

The primary clustering division formula of the machine base loose feature sample set is:

The definition formula of the average distance within the category and the average distance between categories is as follows:

The establishment formula of the initial clustering optimization criterion for the loose feature samples of the machine base is:

The calculation formula of the secondary clustering identification of the loose feature sample set of the machine base is:

references

[1] Du Zhe, Liu Sanyang, Qi Xiaogang. A Fuzzy Support Vector Machine with a New Membership Function. Journal of System Simulation, 2009,21(7):1901-1903.

[2] Ding Shifei, Qi Bingjuan, Tan Hongyan. A Review of Support Vector Machine Theory and Algorithms. Journal of University of Electronic Science and Technology of China, 2011,40(1):2-10.

[3] Zhang Biao. Analysis and Research on Feature Selection Algorithms in Text Classification. Hefei: Master's Degree Thesis of University of Science and Technology of China, 2010.

The above-mentioned technical solutions only reflect the preferred technical solutions of the technical solutions of the present invention, and some changes that those skilled in the art may make to certain parts reflect the principles of the present invention and fall within the protection scope of the present invention.

Claims (7)

1. an effective feature sample identification technology for enhancing classifier generalization performance, it is characterized in that, the method comprises the steps: 1) Establishment of classifier generalization performance evaluation index; 2) Construction of fuzzy clustering criteria; 3) One-time clustering and division of feature sample set; 4) The definition of the average distance within a class and the average distance between classes; 5) Establishment of initial clustering optimization criteria; 6) Secondary clustering identification of feature sample set. 2. according to claim 1, face the effective feature sample identification technology of enhanced classifier generalization performance, it is characterized in that, the establishment calculation formula of described classifier generalization performance evaluation index is: R(w)＝ _Remp (w)+Φ(h/l), h≤min([r ² a ² ],n)+1. where Φ(·) is the confidence risk function, h is the VC dimension of the classification function, and l is the number of training samples. It can be seen that the real risk R(w) is composed of two parts: the empirical risk _Remp (w) and the confidence risk Φ(·). [·] indicates the integer part. r contains all high-dimensional space mapping points The minimum hypersphere radius of . 3. according to claim 1, face the effective feature sample identification technology of enhanced classifier generalization performance, it is characterized in that, the construction computing formula of described fuzzy clustering criterion is: $\mathrm{<span\; class="notranslate">\; min</span>}{\mathrm{<span\; class="notranslate">\; J</span>}}_{\mathrm{<span\; class="notranslate">\; f</span>}\mathrm{<span\; class="notranslate">\; C</span>}\mathrm{<span\; class="notranslate">\; m</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; u</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; V</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\underset{\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; no</span>}}{<span\; class="notranslate">\; \&Sigma;</span>}}\underset{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; c</span>}}{<span\; class="notranslate">\; \&Sigma;</span>}}{<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; u</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}\mathrm{<span\; class="notranslate">\; k</span>}}<span\; class="notranslate">\; )</span>}^{\mathrm{<span\; class="notranslate">\; m</span>}}{<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; d</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}\mathrm{<span\; class="notranslate">\; k</span>}}<span\; class="notranslate">\; )</span>}^{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; .</span>$ In the formula, d _ik ＝||x _k -v _i| | is the distance between the sample x _k and the cluster center v _i , which is generally measured by Euclidean distance. m is the fuzzy weighting index, usually m=2. J _FCM (U, V) represents the sum of squares of weighted distances from various samples to the cluster center, and the weight is the mth power of the membership degree of sample x _k to the i-th class. 4. according to claim 1, the effective feature sample identification technology facing enhanced classifier generalization performance, it is characterized in that, described feature sample set primary clustering division formula is: ${\mathrm{<span\; class="notranslate">\; v</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}^{\mathrm{<span\; class="notranslate">\; l</span>}}<span\; class="notranslate">\; =</span>\underset{\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; no</span>}}{<span\; class="notranslate">\; \&Sigma;</span>}}{<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; u</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}\mathrm{<span\; class="notranslate">\; k</span>}}^{\mathrm{<span\; class="notranslate">\; l</span>}}<span\; class="notranslate">\; )</span>}^{\mathrm{<span\; class="notranslate">\; m</span>}}{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; k</span>}}<span\; class="notranslate">\; /</span>\underset{\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; no</span>}}{<span\; class="notranslate">\; \&Sigma;</span>}}{<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; u</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}\mathrm{<span\; class="notranslate">\; k</span>}}^{\mathrm{<span\; class="notranslate">\; l</span>}}<span\; class="notranslate">\; )</span>}^{\mathrm{<span\; class="notranslate">\; m</span>}}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; K</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; c</span>}<span\; class="notranslate">\; ,</span>$ ${\mathrm{<span\; class="notranslate">\; u</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}\mathrm{<span\; class="notranslate">\; k</span>}}^{\mathrm{<span\; class="notranslate">\; l</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; 1</span>}}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; /</span>\underset{\mathrm{<span\; class="notranslate">\; j</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; c</span>}}{<span\; class="notranslate">\; \&Sigma;</span>}}{<span\; class="notranslate">\; (</span>\frac{{\mathrm{<span\; class="notranslate">\; d</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}\mathrm{<span\; class="notranslate">\; k</span>}}}{{\mathrm{<span\; class="notranslate">\; d</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}\mathrm{<span\; class="notranslate">\; k</span>}}}<span\; class="notranslate">\; )</span>}^{\frac{\mathrm{<span\; class="notranslate">\; 2</span>}}{\mathrm{<span\; class="notranslate">\; m</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; 1</span>}}}<span\; class="notranslate">\; ,</span><span\; class="notranslate">\; \&ForAll;</span>\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; ,</span><span\; class="notranslate">\; \&ForAll;</span>\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; .</span>$ In the formula, the number of clusters c, the fuzzy weight index m and the initial membership degree matrix U ⁰ are set, and the number of iteration steps l=0. For a given stop value ε>0, iterative calculation until max{|u _ik ^l -u _ik ^l-1 |}<ε, the algorithm terminates; otherwise l=l+1, the algorithm continues to execute. 5. according to claim 1, face the effective feature sample identification technology of enhanced classifier generalization performance, it is characterized in that, the definition calculation formula of average distance in described class and average distance between classes is: $\left\{\begin{array}{c}{\mathrm{<span\; class="notranslate">\; \&delta;</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}\mathrm{<span\; class="notranslate">\; no</span>}\mathrm{<span\; class="notranslate">\; no</span>}\mathrm{<span\; class="notranslate">\; e</span>}\mathrm{<span\; class="notranslate">\; r</span>}}<span\; class="notranslate">\; =</span>\underset{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{{\mathrm{<span\; class="notranslate">\; no</span>}}_{\mathrm{<span\; class="notranslate">\; o</span>}}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}\underset{\mathrm{<span\; class="notranslate">\; j</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{{\mathrm{<span\; class="notranslate">\; no</span>}}_{\mathrm{<span\; class="notranslate">\; o</span>}}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}<span\; class="notranslate">\; |</span><span\; class="notranslate">\; |</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}<span\; class="notranslate">\; |</span><span\; class="notranslate">\; |</span><span\; class="notranslate">\; /</span>{\mathrm{<span\; class="notranslate">\; C</span>}}_{{\mathrm{<span\; class="notranslate">\; no</span>}}_{\mathrm{<span\; class="notranslate">\; o</span>}}}^{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; ,</span>\\ {\mathrm{<span\; class="notranslate">\; \&delta;</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}\mathrm{<span\; class="notranslate">\; no</span>}\mathrm{<span\; class="notranslate">\; t</span>}\mathrm{<span\; class="notranslate">\; e</span>}\mathrm{<span\; class="notranslate">\; r</span>}}<span\; class="notranslate">\; =</span>\underset{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; c</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}\underset{\mathrm{<span\; class="notranslate">\; j</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; c</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}<span\; class="notranslate">\; |</span><span\; class="notranslate">\; |</span>{\mathrm{<span\; class="notranslate">\; v</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; v</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}<span\; class="notranslate">\; |</span><span\; class="notranslate">\; |</span><span\; class="notranslate">\; /</span>{\mathrm{<span\; class="notranslate">\; C</span>}}_{\mathrm{<span\; class="notranslate">\; c</span>}}^{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; ,</span><span\; class="notranslate">\; .</span>\end{array}\right.$ In the formula is the number of pairwise combinations of data samples in the cluster {X _O }. v _i and v _j are the centers of the i-th cluster {X _i } and the j-th cluster {X _j } respectively. C _c ² is the combination number of pairwise combinations of c cluster numbers. 6. according to claim 1, face the effective feature sample identification technology of enhanced classifier generalization performance, it is characterized in that, the establishment computing formula of described initial clustering optimization criterion is: $<span\; class="notranslate">\; \{</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; f</span>}}<span\; class="notranslate">\; \}</span><span\; class="notranslate">\; \&DoubleLeftArrow;</span><span\; class="notranslate">\; \{</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; o</span>}}<span\; class="notranslate">\; \}</span><span\; class="notranslate">\; ,</span>$ $\begin{array}{cc}\mathrm{<span\; class="notranslate">\; the\; s</span>}<span\; class="notranslate">\; .</span>\mathrm{<span\; class="notranslate">\; t</span>}<span\; class="notranslate">\; .</span>& \underset{\mathrm{<span\; class="notranslate">\; o</span>}}{\mathrm{<span\; class="notranslate">\; m</span>}\mathrm{<span\; class="notranslate">\; a</span>}\mathrm{<span\; class="notranslate">\; x</span>}}<span\; class="notranslate">\; \&lsqb;</span>{\mathrm{<span\; class="notranslate">\; no</span>}}_{\mathrm{<span\; class="notranslate">\; o</span>}}{\mathrm{<span\; class="notranslate">\; C</span>}}_{{\mathrm{<span\; class="notranslate">\; no</span>}}_{\mathrm{<span\; class="notranslate">\; o</span>}}}^{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; /</span>\underset{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{{\mathrm{<span\; class="notranslate">\; no</span>}}_{\mathrm{<span\; class="notranslate">\; o</span>}}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}\underset{\mathrm{<span\; class="notranslate">\; j</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{{\mathrm{<span\; class="notranslate">\; no</span>}}_{\mathrm{<span\; class="notranslate">\; o</span>}}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}<span\; class="notranslate">\; |</span><span\; class="notranslate">\; |</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}<span\; class="notranslate">\; |</span><span\; class="notranslate">\; |</span><span\; class="notranslate">\; \&rsqb;</span>\end{array}<span\; class="notranslate">\; .</span>$ $<span\; class="notranslate">\; \{</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; no</span>}}<span\; class="notranslate">\; \}</span><span\; class="notranslate">\; \&DoubleLeftArrow;</span><span\; class="notranslate">\; \{</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; o</span>}}<span\; class="notranslate">\; \}</span><span\; class="notranslate">\; ,</span>$ $\begin{array}{cc}\mathrm{<span\; class="notranslate">\; the\; s</span>}<span\; class="notranslate">\; .</span>\mathrm{<span\; class="notranslate">\; t</span>}<span\; class="notranslate">\; .</span>& \underset{\mathrm{<span\; class="notranslate">\; o</span>}}{\mathrm{<span\; class="notranslate">\; min</span>}}<span\; class="notranslate">\; \&lsqb;</span>{\mathrm{<span\; class="notranslate">\; no</span>}}_{\mathrm{<span\; class="notranslate">\; o</span>}}{\mathrm{<span\; class="notranslate">\; C</span>}}_{{\mathrm{<span\; class="notranslate">\; no</span>}}_{\mathrm{<span\; class="notranslate">\; o</span>}}}^{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; /</span>\underset{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{{\mathrm{<span\; class="notranslate">\; no</span>}}_{\mathrm{<span\; class="notranslate">\; o</span>}}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}\underset{\mathrm{<span\; class="notranslate">\; j</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{{\mathrm{<span\; class="notranslate">\; no</span>}}_{\mathrm{<span\; class="notranslate">\; o</span>}}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}<span\; class="notranslate">\; |</span><span\; class="notranslate">\; |</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}<span\; class="notranslate">\; |</span><span\; class="notranslate">\; |</span><span\; class="notranslate">\; \&rsqb;</span>\end{array}<span\; class="notranslate">\; .</span>$ In the formula, {X _f } is an initial effective cluster with a capacity of n _f and a center of v _f which is composed of effective feature samples contained in c clusters (usually c≥3), and {X _n } is mainly composed of The initial invalid cluster {X _n } with capacity n _n and center v _n is composed of noise or wild points. 7. according to claim 1, the effective feature sample identification technology facing enhanced classifier generalization performance, it is characterized in that, described feature sample set secondary clustering recognition calculation formula is: $\begin{array}{cccc}\mathrm{<span\; class="notranslate">\; i</span>}\mathrm{<span\; class="notranslate">\; f</span>}& <span\; class="notranslate">\; |</span><span\; class="notranslate">\; |</span>{\mathrm{<span\; class="notranslate">\; v</span>}}_{\mathrm{<span\; class="notranslate">\; d</span>}}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; v</span>}}_{\mathrm{<span\; class="notranslate">\; f</span>}}<span\; class="notranslate">\; |</span><span\; class="notranslate">\; |</span><span\; class="notranslate">\; <</span>{\mathrm{<span\; class="notranslate">\; \&delta;</span>}}_{\mathrm{<span\; class="notranslate">\; int</span>}\mathrm{<span\; class="notranslate">\; e</span>}\mathrm{<span\; class="notranslate">\; r</span>}}<span\; class="notranslate">\; ,</span>& \mathrm{<span\; class="notranslate">\; t</span>}\mathrm{<span\; class="notranslate">\; h</span>}\mathrm{<span\; class="notranslate">\; e</span>}\mathrm{<span\; class="notranslate">\; no</span>}& <span\; class="notranslate">\; \{</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; d</span>}}<span\; class="notranslate">\; \}</span><span\; class="notranslate">\; \&DoubleRightArrow;</span><span\; class="notranslate">\; \{</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; f</span>}}<span\; class="notranslate">\; \}</span>\end{array}<span\; class="notranslate">\; .</span>$ $\begin{array}{cccc}\mathrm{<span\; class="notranslate">\; e</span>}\mathrm{<span\; class="notranslate">\; l</span>}\mathrm{<span\; class="notranslate">\; the\; s</span>}\mathrm{<span\; class="notranslate">\; e</span>}& <span\; class="notranslate">\; \{</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; the\; s</span>}}<span\; class="notranslate">\; \}</span><span\; class="notranslate">\; \&DoubleRightArrow;</span><span\; class="notranslate">\; \{</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; f</span>}}<span\; class="notranslate">\; \}</span><span\; class="notranslate">\; ,</span>& \mathrm{<span\; class="notranslate">\; a</span>}\mathrm{<span\; class="notranslate">\; no</span>}\mathrm{<span\; class="notranslate">\; d</span>}& <span\; class="notranslate">\; \{</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; t</span>}}<span\; class="notranslate">\; \}</span><span\; class="notranslate">\; \&DoubleRightArrow;</span><span\; class="notranslate">\; \{</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; no</span>}}<span\; class="notranslate">\; \}</span>\end{array}<span\; class="notranslate">\; ,</span>$ $\begin{array}{cc}\mathrm{<span\; class="notranslate">\; the\; s</span>}<span\; class="notranslate">\; .</span>\mathrm{<span\; class="notranslate">\; t</span>}<span\; class="notranslate">\; .</span>& \underset{\mathrm{<span\; class="notranslate">\; the\; s</span>}}{\mathrm{<span\; class="notranslate">\; min</span>}}<span\; class="notranslate">\; \&lsqb;</span>\underset{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{{\mathrm{<span\; class="notranslate">\; no</span>}}_{\mathrm{<span\; class="notranslate">\; the\; s</span>}}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}\underset{\mathrm{<span\; class="notranslate">\; j</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{{\mathrm{<span\; class="notranslate">\; no</span>}}_{\mathrm{<span\; class="notranslate">\; the\; s</span>}}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}<span\; class="notranslate">\; |</span><span\; class="notranslate">\; |</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}<span\; class="notranslate">\; |</span><span\; class="notranslate">\; |</span><span\; class="notranslate">\; |</span><span\; class="notranslate">\; |</span>{\mathrm{<span\; class="notranslate">\; v</span>}}_{\mathrm{<span\; class="notranslate">\; the\; s</span>}}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; v</span>}}_{\mathrm{<span\; class="notranslate">\; f</span>}}<span\; class="notranslate">\; |</span><span\; class="notranslate">\; |</span><span\; class="notranslate">\; /</span>{\mathrm{<span\; class="notranslate">\; C</span>}}_{{\mathrm{<span\; class="notranslate">\; no</span>}}_{\mathrm{<span\; class="notranslate">\; the\; s</span>}}}^{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; \&rsqb;</span>\end{array}<span\; class="notranslate">\; .</span>$ In the formula, {X _s } is a combined subset with capacity n _s and center v _s extracted from {X _d }, which satisfies the minimization criteria. After effective sample extraction, the remaining data samples in {X _d } form a subset {X _t }, which is merged into the invalid cluster {X _n } composed of noise (wild) points. Equation (18) implements the secondary division process of invalid clusters, where _xi is the data sample in the invalid cluster {X _n } formed after the main division. X _near is the data sample closest to the center v _n of the invalid cluster {X _n } in the valid cluster {X _f }.