TW200818060A - Clustering system, and defect kind judging device - Google Patents

Abstract

Provided is a clustering system capable of classifying target data more rapidly and precisely than the examples of the prior art. The clustering system classifies input data into each of clusters formed of the populations of learning data, in terms of featuring quantities owned by the input data. The clustering system comprises a featuring set storage unit stored with featuring quantity sets or a combination of featuring quantities to be used for the classifications, in a manner to correspond to the individual clusters, a featuring quantity extracting unit for extracting a preset featuring quantity from the input data, a distance calculating unit for calculating and outputting the distances between the center of the population of the individual clusters and the input data, individually as set distances for each featuring quantity set corresponding to each cluster, on the basis of the featuring quantities contained in the featuring quantity set, and a rank extracting unit for arraying the individual set distances in insequence.

Description

200818060 IX. Description of the Invention [Technical Field of the Invention] The present invention relates to a partial image of a defective portion in a portrait of a detection object, from which a feature signal of a defect is extracted, and the defect is given by the type of the defect. Classification cluster system, defect type determination device. [Prior Art] Clustering techniques such as those generated by the Mahalanobis' generalized distance have been conventionally performed. That is, the cluster processing is performed by classifying whether the unknown data belongs to a cluster of the parent group as the prior learning. For example, by judging the size of the Mahalanobis distance with respect to the majority cluster, it is determined which cluster of the parent group the unknown material belongs to (for example, refer to Patent Document 1). Furthermore, in order to efficiently calculate the above distance, the processing of the cluster is performed by selecting a plurality of feature quantities. Furthermore, based on the votes obtained by the majority of the classifiers, the method of determining the cluster to which the unknown data belongs is also generally visible, using the recognition result of the output of the different sensors, or on a portrait. The recognition result of the unknown data is recognized in different fields (for example, 'refer to Patent Document 2). By the above-mentioned clustering method, the disease diagnosis is performed on the parameters obtained according to the results of the blood test, that is, the cluster in which the disease belongs. 'Every two clusters are set in a majority cluster as a combination, and all combinations are judged. Whether or not the combined test materials are similar to any cluster's by the total of the number of the determinations -5 - 200818060, the method of classifying into the determined number of clusters is determined (for example, refer to Patent Document 3). When each defect attached to the LCD glass substrate is classified into each defect type set in advance, the optimization of each feature amount used for classification is performed corresponding to the identification at the time of classification to correspond to the optimization. In the manner, weighting is performed for each feature amount, and the cluster analysis to which cluster belongs is determined using the optimized feature amount (for example, Patent Document 4). [Patent Document 1] Japanese Patent Laid-Open Publication No. 2001-56861 Patent Document 3: Japanese Patent Application Laid-Open No. Hei 7- 1 0 5 1 6 6 [Problem to be Solved by the Invention] However, the cluster analysis shown in Patent Document 3 does not perform optimization of each combination, and does not effectively use the discrimination. The feature quantity of the material, and when the cluster to be discriminated is increased, there is a problem that the number of combinations is large and the time taken for the determination process is increased. Furthermore, the cluster analysis shown in Patent Document 4 is based on the determination rate, and the feature quantity is improved with the weighted discrimination accuracy, but the feature quantity of each cluster is not optimized, and the above patent document 3 is the same, the feature quantity cannot be effectively utilized, so it is impossible to perform high-precision classification. In view of the fact that the author of the present invention provides a feature quantity -6 - 200818060 that can be extracted from the classification object data classified as an object belonging to the cluster, the present invention uses a higher speed and higher precision than the conventional example. The object data is classified, for example, a defect system located on a glass surface is classified into a cluster system corresponding to a cluster of defect types, and a defect type determining device. [Means for Solving the Problem] In order to solve the above problems, the present invention differs from the conventional example in which the classification target data and the clusters are calculated by the same type of feature quantity, and the classification point is determined. The cluster setting can obtain a set of the feature quantities of the differences between the clusters, and obtain the distances between the clusters with different feature quantities, so that the classification is performed with higher precision than in the past. Since the set of the above feature quantities is executed based on the characteristics of the learning materials belonging to each cluster, it is composed of feature quantities that can be distinguished from other clusters. That is, the present invention adopts the following constitution. The cluster system of the present invention belongs to a cluster in which input data is classified into a cluster formed by a population of learning data by a parameter of the input data. The system is characterized in that: the feature quantity set storage unit 'corresponds to each of the clusters, and stores a parameter set of a combination of feature quantities used for classification; the feature quantity extraction unit extracts the prior information from the input data The set feature quantity; the distance calculation unit calculates the distance between the center of the parent group of each cluster and the input data based on the feature quantity 'accepted by the feature quantity set for each feature quantity set of each cluster set And outputting each set distance; and the order extracting unit, and the respective set distances are arranged from small to large. 200818060 The optimal clustering system of the present invention has a plurality of settings in each cluster. The optimal clustering system of the present invention further includes a cluster classifying section that classifies the classification criteria of the clusters by inputting the input data of the set distances of the set distances obtained in each of the feature amount sets. The mode detects which cluster the input data belongs to. In the optimal clustering system of the present invention, the cluster classification unit detects, by the rank of the set distance, which cluster the input data belongs to, and sets the rank as a cluster of the upper set distance as the input data. It is detected by the cluster it belongs to. In the optimal clustering system of the present invention, the cluster classification unit has a threshold 数量 which is a higher number with respect to the rank, and if the upper cluster is above the threshold ,, it is detected as a cluster to which the input data belongs. In the optimum clustering system of the present invention, the distance calculating unit multiplies the correction coefficient set corresponding to the feature amount with respect to the collective distance, and normalizes the collective distance between the feature amount sets. The optimal clustering system of the present invention further has a feature quantity set cooperation unit for creating a feature quantity set of each cluster, and the feature quantity set cooperation part is a mother of each cluster for each combination of a plurality of combinations of feature amounts. The average of the Group's learning materials is taken as the origin, and the average distance between the origin and the learning materials of the parent group of the other clusters is obtained, and will be the combination of the maximum average 特征 characteristic amount, which is selected for A collection of feature quantities that each cluster recognizes from other clusters. The defect type determination device according to the present invention is characterized in that, in any one of the cluster systems described in the above-mentioned -8-200818060, the input data is an image data of a product defect, and the image is represented by a feature amount indicating the defect. Defects in the data are classified according to the type of defect. In the optimum defect type determining device of the present invention, the product is a glass article, and the defects of the glass article are classified according to the type of the defect. The defect detecting device of the present invention is characterized in that the defect type determining means described above is provided for detecting the type of defects of the product. The manufacturing state judging device according to the present invention is characterized in that the defect type judging device as described above is provided, and the defect classification of the product is executed, and the occurrence of the defect in the process is detected based on the correspondence with the cause of occurrence of the category. The cause. An optimum manufacturing state judging device according to the present invention is characterized in that the cluster system described above is provided, and the input data is a feature amount indicating a manufacturing condition in a process of the product, and is in accordance with a manufacturing state of each process of the process. The feature quantities are classified. In the optimum manufacturing state judging device of the present invention, the product is a glass article, and the feature amount in the process of the glass article is classified according to the manufacturing state of each process of the process. The manufacturing state detecting device according to the present invention is characterized in that the manufacturing state determining device described above is provided to detect the type of the manufacturing state in each of the processes of the product process. The product manufacturing management apparatus according to the present invention is characterized in that the manufacturing state determining means described above is provided, and the type of the manufacturing state in each of the processes for detecting the product process is executed, and the control item corresponding to the category is -9- 200818060 Process control in process engineering. [Effects of the Invention] As described above, according to the present invention, for each cluster of classification points, a combination of the best feature amounts that are farther away from the other clusters is set in advance from the majority of the feature amounts of the classification target data. Calculating the distance between the classification object data and each cluster, because the classification object data is classified to the cluster where the calculated distance is the smallest, the classification object data can be classified more correctly to the corresponding cluster than the conventional method. Furthermore, if the above combination is set in each cluster majority by the present invention, the distances of the calculation results of the full cluster and the classification object data are arranged in ascending order, and the classification object data is classified into a pre-set number of upper clusters. With the largest number of clusters included, it is possible to perform high-precision classifications compared to the past. [Embodiment] The cluster system of the present invention relates to a cluster system that classifies input data of a classification object into respective clusters formed by using the learning material as a parent group by the feature quantity of the input data, and has a feature quantity. The set memory unit is configured to collect a feature amount set corresponding to a combination of feature amounts used for classification in accordance with each of the clusters, and extract a feature amount and a distance from the input data based on the feature amount set in which the feature amount extracting unit is set in advance. The calculation unit calculates the distance between the parent group and the input data according to the feature quantity included in the feature quantity set, and calculates the distance from the input data to the set -10- 200818060. The set distances are arranged in ascending order, and the clustering is performed corresponding to the order of the columns. [First Embodiment] Hereinafter, a cluster system according to a first embodiment of the present invention will be described with reference to the drawings. Fig. 1 is a block diagram showing a configuration example of a cluster system according to the embodiment. The present embodiment is a feature set integration unit 1, a feature amount extracting unit 2, a distance calculating unit 3, a feature amount set storage unit 4, and a cluster database 5, as shown in Fig. 1. In the feature amount set storage unit 4, a feature amount set in which a combination of feature amounts of the classification target data is set in each cluster is stored in association with the identification information of each cluster. For example, when the classification target data is the set of feature quantities {a, b, c, d}, it is set as the feature quantity set u, b], [a, b, c, d], [c], etc. of each cluster. A combination of feature quantities of the types. In the following description, 'any combination of feature amounts, a plurality of combinations (in the above example, two of the sets, three feature amounts) is defined as "combination of feature quantities" from the set of feature amounts described above. . When clusters A, B, and C are set as clusters of classification locations, the feature quantity sets corresponding to the clusters are used to be classified into clusters in advance; ^ learning materials are used to find distances between clusters and other clusters. The combination of the characteristics of the big features is stored in the feature quantity set storage unit 4. For example, the set of feature quantities set for cluster A is a vector consisting of the average 値 of the feature quantities of the learning materials belonging to bundle _A, and the learning materials of genus -11 - 200818060 other clusters B and C. The distance formed by the average 値 of each feature amount is set as the combination of the largest feature amounts. Furthermore, the classification object data and the learning materials of the parent group in each cluster are composed of the same set of feature quantities. When the distance between the clusters is calculated from the input classification target data, the feature amount extracting unit 2 reads out the feature amount storage unit 4 corresponding to the cluster to be calculated from the feature amount set storage unit, and the feature amount of the self-classification target data The feature amount corresponding to the feature amount combination is extracted, and the extracted feature amount is output to the distance calculation unit 3. The distance calculation unit 3 is a self-cluster database 5, and reads, as a key, identification information of a cluster to be calculated, and reads a vector composed of an average value of each feature amount of the learning material of the cluster to be calculated, according to the cluster. The feature quantity set calculates the distance between the vector composed of the feature quantity extracted from the classification target data and the vector formed by the average value of each feature quantity of the learning material (the center of gravity vector indicating the position of the center of gravity of most of the learning materials in the cluster) When the distance calculation unit 3 performs the calculation of the distance, the distance calculation unit 3 normalizes the number of feature quantities by the difference in the data unit of the feature quantity. Therefore, the feature quantity v of the classification target data is expressed by the following formula (1). (i) Perform formalization. V(i) = (v(i)-avg.(i))/std.(i)...(l) Here, v ( i ) is the feature quantity, and avg· ( i ) is the cluster of the calculation object The average of the feature quantities in the learning materials, std. ( i ) is the standard deviation of the feature quantities in the learning data in the cluster of calculations ( -12- 200818060 standardized deviation ), v(i) is normalized The feature quantity. Therefore, when calculating the distance, the distance calculating unit 3 must integrate each feature amount and perform normalization of each feature amount. Further, the distance calculating unit 3 calculates each feature amount used for calculating the distance in the classification target data, and performs the above-described normalization processing using the average 値 and standard deviation of the respective feature amounts of the learning materials. Furthermore, as the distance, a standardized Euclidean distance, a Mahalanobis distance, a Minkowski ydistance, etc., using the above-described standardized feature quantities are employed. Any of them. Here, when the Mahalanobis distance is used, the Mahalanobis' squared distance MHD shell [] is obtained by the following formula (2). MHD = (l/n) · (VTR~1V)...(2)

The feature quantity v(i) of each element V in the rank V in the above formula (2) relative to the multiple elements of the unknown data is the average 値avg of the feature quantity by the learning data in the cluster.  (i) and standard deviation std.  (i) The feature quantity obtained by the above formula (1). η is a degree of freedom, and in the present embodiment, the number of feature quantities indicating the number of feature quantities in the feature amount set (described later). Thereby, the Mahalanobis squared distance is the number 加 of the difference of the η transformed characteristic quantities, and the unit distance of the parent group average becomes 1 by (Marananobis squared distance) / η. Furthermore, VT is the transposed rank of the rank V of the feature quantity v(i), and R·1 v is the correlatione matrix between the feature quantities in the learning data in the cluster R -13- The retrograde column of 200818060. The feature amount set cooperation unit 1 calculates a feature amount set used when the distance calculation unit calculates a distance between the classification target data and each cluster in each cluster, and writes the calculation result to each cluster identification information, and writes the feature amount to the feature amount. The memory unit 4 is stored and memorized. When calculating the feature quantity set, the feature quantity set cooperation part 1 is a barycentric vctor of the learning material belonging to the object cluster of the generated feature quantity set for each cluster, and belongs to all other than the object cluster. The distance of the center of gravity vector of the learning data of the cluster is used as a reference, and 値λ of the discriminant criterion is calculated by the following equation (3). Hereinafter, the combination of the feature amounts will be described as a feature amount set. λ = ω〇ωί(μ〇-μΐ)2/(ω〇σ〇2 + ωίσΐ2). . . (3) In the above (3), μ is a centroid vector consisting of the average 特征 of the feature quantities in the feature quantity set of the learning material belonging to the target cluster (the parent group in the cluster). σ is the standard deviation of the vector generated by the feature quantities of the learning materials belonging to the parent group within the cluster. (1) The ratio of the number of learning materials belonging to the parent group in the cluster to the number of learning materials belonging to the full cluster. Again, μ. The center of gravity vector consisting of the average of the feature quantities in the feature quantity set of the learning material (object cluster outer group) belonging to the cluster outside the cluster. σ. It is the standard deviation of the vector generated by the feature quantity of the learning material belonging to the external cluster of the object cluster. ω. The ratio of the number of learning materials belonging to the cluster outer group to the number of learning materials belonging to the full cluster. Here, in the formula (3), (μ.,.) can be used even if the log (logarithm) and the square root number are used. Furthermore, when calculating each vector, the feature quantity set cooperates as -14- 200818060. Section 1 calculates the feature quantity normalized for each feature quantity by the equation (1). Further, even if the inherent 値 of the ratios ωι and ω◦ are set in advance, the number of separations becomes large. Then, the feature amount set cooperation unit 1 calculates the discrimination criterion 値λ which is discriminated from the other clusters by using the above equation (3) for each of the target clusters and for any or all of the feature quantities constituting the learning material. The ranking list of the discrimination reference 値 λ is outputted by arranging the calculated discrimination reference 値 λ ' in descending order. Here, the feature quantity set cooperation unit 1 is a combination of the feature quantities corresponding to the maximum discrimination reference 値λ as the feature quantity set of the target cluster, and the discrimination reference 値λ, and corresponding to the identification information of the cluster, and the memory The feature quantity set storage unit 4. The determination criterion 値λ is determined as shown in Fig. 2(a). When the feature amount set cooperation unit 1 sets the feature quantity set of each cluster, the feature quantities of the learning data and the classification target data are a and b. When four of c, d and d are used, the discriminant reference 値λ of all of the four feature quantities, all of them, and one of them is calculated. Then, the feature quantity set cooperation unit 1 selects the highest number 値, for example, the combination of the feature quantities b and c is selected in Fig. 2(a). Further, in the method of the other criterion 値λ, there is a BSS method as described in FIG. 2(b), that is, a discrimination criterion 値λ for calculating all the feature amounts η included in the set of classification target data. Next, the combination of η_1 is extracted from the set of θ of the feature quantities, and the discrimination criterion 値λ is calculated. Then, the combination of the maximum 値 -15-200818060 is selected from the η-1 discriminant reference 値λ, and the reference 値λ is calculated next time from the n-l feature amounts to n-2 all combinations. In this way, even if the feature quantity is reduced by the self-collection by the order, the set of the feature quantity of the self-reduction is selected, and the combination is reduced by one, and the criterion 値λ is calculated, and the combination can be determined by the less feature quantity. In this manner, the feature quantity set cooperation unit 1 may be configured. Furthermore, in the method of the other criterion 値λ, the FSS method as described in FIG. 2(c) is a feature quantity η one by one read feature included in the set of self-classification target data. For all types, the discrimination criterion 値λ of each feature amount is calculated, and the feature amount having the largest discrimination reference 値 is selected. Next, a combination of the feature amounts and the two feature amounts of the other feature amounts is generated, and the discrimination reference 値λ with respect to each combination is calculated. Then, the combination having the largest discriminant reference 値 is selected from the combination. Next, a combination of the three feature quantities of the combination and the feature not included in the combination is generated, and each discrimination criterion 値λ is generated. In this way, even if the feature quantity having the largest discriminant reference 値λ is selected in the order from the combination of the feature quantities before the order, the feature quantity of the combination is increased by i with respect to the combination, and the feature quantity that does not exist in the combination is calculated. The discriminant reference 値λ of the combined feature quantity is calculated, and the discriminant reference 値λ of the combination of the feature quantities which are not present in the feature quantity of the combination is calculated, and finally, all combinations of the discrimination reference 値λ are calculated, and the criterion is determined. It is also possible to form the feature quantity set cooperation unit 1 in such a manner that the largest combination of 値λ is selected as the feature quantity set. Next, by discriminating the reference 値λ, the validity of the selection of the feature quantity set used in the cluster is represented by the third and fourth figures. -16- 200818060 Fig. 3 is a combination of the feature quantities a and h combined with the feature quantities a, b, c, d, and the combination of the feature quantities d and e, as a selection The combination of the feature quantity sets, from these combinations, in the cluster 1 , the cluster 2 , and the cluster 3 , a set of feature quantities having high classification characteristics is selected as compared with the conventional example. In Fig. 3, μΐ corresponds to μ; and μ2 corresponds to μ. , σ ΐ corresponds to σ i , and σ 2 corresponds to σ. , ω 1 corresponds to ω 1, and ω 2 corresponds to ω 〇. Wherein, in the above combination, the criterion 値λ is the largest combination of the feature quantities a and h', and the combination is used for the separation of the cluster 1 and the other clusters, and the cluster 1 is confirmed by the fourth graph and Classification results for clusters other than clusters (clusters 2 and 3). In Fig. 4, the horizontal axis represents the number of log of the Mahalanobis distance calculated using the combination of the feature quantities, and the vertical axis represents the number of the separated object data (histogram) having the corresponding number 値. Here, the number of the horizontal axis is 値1. 4 means that the number of logs of Mahalanobis is less than 1. 4 must and 1. 2 or more (1. 4 on the left side 値). The numbers on the other horizontal axes are also the same. Furthermore, in Figure 4, it indicates 1. 4S means that I·4 or more. The Mahalanobis distance in Fig. 4 is calculated using the feature quantity set corresponding to cluster 1, and the classification object data belonging to cluster 1 and other clusters are calculated. Fig. 4 (a) is a use feature An example of calculating the Mahalanobis distance by the combination of the quantities a and g, and (b) is an example of calculating the Mahalanobis distance using the combination of the feature quantities & and h, FIG. 4 ( C) An example of calculating the Mahalanobis distance using a combination of the characteristic quantities d and e of the 17-200818060. When viewing the histogram in Fig. 4, when the number of discriminations 値λ is large, it is known that the classification of cluster 1 and other clusters is well performed. Next, the operation of the cluster system according to the embodiment of Fig. 1 will be described with reference to Figs. 5 and 6 . Fig. 5 is a flowchart showing an operation example of the feature quantity set cooperation unit i of the cluster according to the first embodiment, and Fig. 6 is a flowchart showing an operation example of the cluster of the classification team image data. In the following description, for example, when the classification target data is a set of the feature amount of the scratch on the glass article, "a: the length of the scratch" which is the feature amount is obtained from the image processing or the measurement result. "b: the area of the scratch", "c: the width of the scratch", "d: the transmittance of the specific region containing the scratched portion", and "e · the reflectance of the specific region containing the scratch". Therefore, the set of feature quantities (hereinafter, the feature quantity set) becomes {a, b, c, d}. Furthermore, in the present embodiment, the distance used for the cluster analysis is calculated as the Mahalanobis distance using the normalized feature amount. Here, the glass article in the present embodiment may be, for example, a plate glass or a glass substrate for a display. A. Feature amount set cooperation processing (corresponding to the flowchart of Fig. 5) The user detects a scratch on the glass, photographs the image to obtain image data, and performs image processing to extract a scratch portion from the image data. The feature amount such as the length measurement is used to collect the feature amount data composed of the above-described feature amount. Then, for each cluster classified by the user in the cause or shape of the scratches, the feature amount is divided into learning materials according to the known cause or the information of the shape, and the learning is performed as a cluster. The parent group of the data, from the non-graphic processing terminal corresponding to the identification information of the cluster, is memorized to the cluster database 5 (step S1). Next, when the feature amount set cooperation unit 1 inputs a control command for generating a feature amount set to each cluster from the processing terminal, the cluster data library 5 reads the identification information of each cluster and reads the parent group of the learning materials. Then, the feature quantity set cooperation unit 1 calculates the average 値 and standard deviation of each feature quantity in the parent group in the cluster for each cluster, and uses the average 値 and the standard deviation to calculate the specifications in each learning material from the formula (1). The amount of features. Next, the feature amount set cooperation unit 1 is for each feature amount of a combination of all the feature amounts included in each feature amount set, and the discrimination reference 値λ is calculated by the equation (3). In this case, the feature quantity set cooperation unit 1 calculates the normalized 値 (center of gravity vector) μι of the vector composed of the feature amounts corresponding to the respective feature amount sets, using the normalized feature quantity of the parent group in the cluster for each cluster. Using the normalized feature quantity of the vector of the cluster external group, the characteristic deviation amount of the vector of the learning material which is composed of the feature quantity of the feature quantity set in the parent group in the cluster is calculated, and the feature corresponding to each feature quantity is calculated. The average 値 (center of gravity vector) μ of the vector formed by the quantity. And a standard deviation σ of a vector of learning materials composed of feature quantities corresponding to a set of feature quantities in the cluster outer group. , the ratio of the number of learning materials of the parent group in the cluster of the total number of learning materials ωι, and the ratio of the number of learning materials of the cluster of the external learning group in the total number of learning materials ω 〇〇-19- 200818060 Then, the feature quantity set The preparation unit 1 makes the standard deviations σ i and σ. , ratio ω i, ω. Cluster, the discriminant of the distance between the special cluster and all the clusters of all combinations of feature quantities. When all the discriminant fiducials 値λ are calculated, the knot 1 is the cluster of each cluster from the largest to the smallest. When 値λ is specified, the feature amount set of the table used for calculating the distance is detected (step S2). Then, the feature quantity set cooperation unit 1 calculates the set of feature amounts R corresponding to each feature quantity set R, and the learning 1 avg· (i) and standard deviation std in the parent group in each cluster.  (i) (In the following, the feature quantity set cooperation unit 1 self-correction coefficient 1 (1/2). The normalization between the correction coefficients λ_(]. Because of the different clusters, some deviations occur, so in order to improve the classification accuracy Standardization. Furthermore, instead of using λ_(] to use log ( λ), or simply using a function, the standard between the set of feature quantities can be performed. Furthermore, in the above formula (3), the feature is The centroid vector μ in the quantity set is used to calculate the outer cluster of the target cluster [using the above-mentioned center of gravity vector μί, μ. , by equation (3), in each eigenvalue set, the discriminant of each quasi-値 λ In the case of the ί bundle, the feature quantity set cooperates to form the discriminant reference 値λ, and the eigenvalue set is attributed to the combination of the clustered feature quantities, and the feature of the correlation coefficient of the calculated feature quantity is calculated. The average value of the quantity 骤, step S 3 ). The above-mentioned discriminant criterion 値λ calculates [/2) to obtain the distance between each feature quantity set cluster and other clusters, and it is necessary to execute the feature quantity set ί/2) as the correction coefficient, even if > -μί) Also, if it is a λ-qualifier Is calculated by either parent objects outside the cluster group, they select the following three categories in the learning materials. -20- 200818060 a.  All the learning materials of the external parent group of the objects in the full learning materials b.  Specific learning materials corresponding to the purpose of the classification in the parent group of the above-mentioned object cluster c.  The collection of the objects in the learning materials used for the feature quantity selection. The learning materials of the external parent group. Here, b. The purpose of the classification is to distinguish it from a clear and conspicuous cluster, and the learning material is the learning material contained in other clusters that are intended to confer the difference. Then, the feature quantity set cooperation unit 1 corresponds to the identification information of each cluster, and sets the feature quantity; the correction coefficient corresponding to the feature quantity set is 1 (1/2) in the present embodiment; the inverse row; the average値avg.  (i) ; and standard deviation std.  (i) Memory is stored as the distance calculation data in the feature amount collecting unit 4. B. Cluster processing (corresponding to the flowchart of Fig. 6) When the classification target data is input, the feature amount extracting unit 2 reads out the feature amount set corresponding to each cluster from the feature amount set storage unit 4 by the identification signals of the clusters. Then, the feature quantity extracting unit 2 extracts the feature quantity from each of the classified object data to each cluster based on the type of the feature quantity in the read feature quantity set, and extracts the extracted feature quantity memory corresponding to each piece of identification information of the cluster. In the internal memory (step S 1 1 ). Next, the distance calculating unit 3 reads out the average 値avg·(i) and the standard deviation std(i) of the feature amount from 21 to 200818060 from the feature amount set storage unit 4, by executing the above formula (2) In the calculation, the feature quantity of the classification target data is normalized, and the feature quantity stored in the internal memory unit is replaced with the normalized feature quantity. Then, the distance calculating unit 3 generates the rank V composed of the elements of V(i) obtained as described above, calculates the transposed row VT of the type V, and sequentially calculates the classification target data and the equation (3). The Mahalanobis distance between the clusters corresponds to the identification information of each cluster, and is stored in the internal memory (step S1 2). Next, the distance calculating unit 3 calculates the Mahalanobis distance of the calculation result, multiplies the correction coefficient λ·(1/2) corresponding to the feature amount set, and obtains the correction coefficient, which is different from the Mahalano ratio. Distance replacement (step S 1 3 ). Furthermore, even when multiplying the correction coefficient, it is possible to calculate the log or square root of the Mahalanobis distance. Then, the distance calculating unit 3 compares the correction distances between the clusters in the internal memory unit (step S1 4), detects the smallest correction distance, and uses the cluster of the identification information corresponding to the correction distance as the classification target data. The cluster to which it belongs, for the cluster library 5, the identification information corresponding to the cluster of the classification points, and the classified object data classified (step S15). [Second Embodiment] In the first embodiment, the feature amount set used in the cluster analysis is described in each cluster type. However, even in the second embodiment described below, each cluster is used. Most of the set feature quantities, -22-200818060 operation corresponds to the Mahalanobis distance of each feature quantity set, and the correction distance is calculated, and the correction distance is arranged in the order of small to large, and the correction is corrected by the specific order of the upper position. The distance may be a cluster to which the classification object belongs, in accordance with the rules set in advance. That is, the distance calculation 3' in the present embodiment is classified into the respective clusters by the classification target data which is set based on the order of the distance between the classification target data acquired in each feature amount set and the distance of each cluster. The rule pattern of the classification criteria detects which cluster the classification object data belongs to. In the first embodiment, the configuration of the second embodiment is the same as that of the first embodiment shown in Fig. 1, and the same reference numerals are given to the respective configurations. Only the operation different from the first embodiment in each configuration will be described using Fig. 7 . In the second embodiment, there is a process of setting the rule mode by the self-learning material. Fig. 7 is a flow chart showing an example of the operation of mode learning with respect to the distance rule of the set rule mode. Fig. 8 and Fig. 9 are flowcharts showing an example of the operation of the cluster in the second embodiment. Furthermore, in the first embodiment, when the feature quantity set is created, the feature quantity set cooperation unit 1 calculates the discrimination reference 値λ' for each of the clusters 'the majority of the feature quantity sets that are the combination of the feature amounts. The set of feature quantities of the majority of the determined criterion 値λ is set as the feature quantity set of each cluster. Further, in the second embodiment, the feature amount set cooperation unit 1 sets the maximum of the feature amount sets corresponding to the combination number of the feature amounts for each cluster or one or a plurality of combinations or all other clusters of the other clusters.値-23- 200818060, the majority discriminant reference 値λ is obtained, and a set of most feature quantities for separating from other clusters is set for each cluster. Then, the feature quantity set cooperation unit 1 calculates the distance calculation data for each feature quantity set, and corresponds to the identification information of the cluster, so that the feature quantity of the majority feature quantity and the distance calculation data of each feature quantity set are memorized in the feature quantity set memory. Department 4. Then, in the seventh drawing, when the learning material is input, the feature amount extracting unit 2 reads out the most feature amount sets corresponding to each cluster from the feature amount set storage unit 4 by the identification signals of the clusters. Then, the feature quantity extracting unit 2 is a category corresponding to the feature quantity in each of the read feature quantity sets, and the feature quantity is extracted from the learning material to each cluster so that each piece of identification information corresponding to the cluster is set for each feature quantity. The extracted feature amount is memorized in the internal memory portion (step S2 1). Next, the distance calculating unit 3 is the self-featured set storage unit 4, and reads out the average 値avg of each feature amount set corresponding to the feature amount.  (i) and standard deviation std.  (i) By performing the calculation of the above formula (2), the feature quantities extracted from the self-learning materials are normalized, and the feature quantities stored in the internal memory are replaced with the normalized feature quantities. Then, the distance calculating unit 3 generates the rank V composed of the elements of V(i) obtained as described above, calculates the transposed row VT of the type V, and sequentially calculates the learning material and each by the equation (3). The Mahalanobis distance between the clusters corresponds to the identification information of each cluster, and each feature quantity set is set to the internal I memory portion (step S22). Next, the distance calculating unit 3 calculates the Mahalano ratio -24-200818060 s distance from the calculation result, and multiplies the correction coefficient λ^ι/2 corresponding to the feature quantity set to obtain a correction coefficient, each of which is matched with Maha. The Ronbis distance replacement (step S23). Then, the distance calculating unit 3 arranges the correction distances between the clusters in the internal memory unit in the order of small to large (the small correction distances are arranged in the upper position), that is, the correction distance of the classification target data is small. The identification information of the clusters is arranged in the order of the upper ranks (step S24). Next, the distance calculating unit 3 detects the identification information of the clusters from the small (upper) to the nth corrected distances, and counts the number of pieces of identification information of each of the n included clusters, that is, performs voting for each cluster. Then, the distance calculating unit 3 is a pattern for detecting the number of counts of the identification information of each cluster of each learning material, and is a rule pattern common to the learning materials included in the same cluster. For example, when η is set to 1 〇, when the learning data for the cluster ' is 'detected as 5 clusters, 3 clusters, and cluster C is 2 counts, set this to Rule i. Furthermore, when it is the learning data of cluster C, when three clusters C are detected, even if cluster A is 7 and cluster B is 0, it is not necessarily common to cluster C. If the number of clusters C is 3 or more, Cluster C can also be set irrespective of the number of counts of other clusters, and this is set to rule 2. Furthermore, 'for the learning data of cluster A, cluster A is the arrangement pattern of the first and second digits from the upper position, even if the number of counts of the cluster β is eight, the cluster can be set irrespective of the number of counts of other clusters. A, this is set to rule 3. As described above, the regularity of the number of counts of each cluster which is classified into the same cluster is detected, and is stored as a pattern table in the identification information of each cluster -25-200818060. Here, even if one rule is set in each cluster, most settings may be made. Further, in the above description, although the distance calculation unit 3 extracts the rule mode, the user may arbitrarily set the number of counts or the rule mode in order to change the classification accuracy of the classification to each cluster. According to different clusters, the characteristics of some cluster feature information are similar to those of other clusters. The correlation between the majority clusters is the number of counts of each cluster or the object pattern from the pattern of the upper ranks, and the classification of the data of the classified objects. In the case where the accuracy of one side is relatively high, this embodiment complements this point. Next, the processing of cluster analysis using the second embodiment of the rules described in the above table will be described using the flowchart of Fig. 8. When the classification target data is input, the feature amount extracting unit 2 reads out the majority of the feature amount sets corresponding to each cluster from the feature amount set storage unit 4 by the identification signals of the clusters. Then, the feature quantity extracting unit 2 is a category corresponding to the feature quantity in each of the read feature quantity sets, and extracts the feature quantity from the classification target data to each cluster so that each piece of identification information corresponding to the cluster is used for each feature quantity. The set memorizes the extracted feature amount in the internal memory (step S3 1). Next, the distance calculating unit 3 is the self-featured set storage unit 4, and reads out the average 値avg·(i) and the standard deviation std of each feature amount set corresponding to the feature amount. (i) By performing the above-described equation (2), the feature quantities extracted from the classification target data are normalized, and the feature amount stored in the internal memory portion is replaced with the normalized feature amount. -26-200818060 Then, the distance calculating unit 3 generates the rank V composed of the elements of V(i) obtained as described above, and calculates the transposed row VT of the type V, which is calculated by the equation (3). The Mahalanobis distance between the learning material and each cluster corresponds to the identification information of each cluster, and is stored in the internal memory portion for each feature amount set (step S32). Next, the distance calculating unit 3 calculates the Mahalanobis distance of the calculation result, multiplies the correction coefficient λ·(1/2) corresponding to the feature amount set, and obtains the correction coefficient, which is different from the Mahalano ratio. The distance is replaced (step S3 3 ). Then, the distance calculating unit 3 arranges the correction distances between the clusters in the internal memory unit in the order of small to large, that is, the identification information in which the clustering information of the classification target data is small is arranged in the order of the upper level ( Step S 3 4 ). After the arrangement, the distance calculating unit 3 detects the identification information of the clusters of the correction distances from the small (upper) to the nth, and counts the number of pieces of identification information of each of the n clusters, that is, performs voting on each cluster. . Next, the distance calculating unit 3 executes a matching process of whether or not the mode (or the mode of arrangement) of the number of the upper bits of each of the classification target data with respect to the number of counts of each cluster exists in the table stored therein (step S35). When the distance calculation unit 3 detects the result of the comparison and the rule pattern matching the target pattern of the classification target data is described in the table, the distance calculation unit 3 determines that the classification target data belongs to the cluster of identification information corresponding to the matching rule. The classification object data is classified in the cluster (step S36). Further, the processing of using the other clusters of the second embodiment described in the above table will be described using the flowchart of Fig. 9. -27- 200818060 In the other cluster processing shown in FIG. 9, the processing of step S31 to step S35 is the same as the processing shown in Fig. 8, and the distance calculating unit 3 is as described in step S35. In general, the rule mode of self-memory in the table is executed, and the object mode of the classified object data is processed. Then, the distance calculation unit 3 detects whether or not the rule pattern matching the target pattern is searched for, and when detecting the rule pattern in which the matching is found, the process proceeds to step S47, and when When it is detected that the rule pattern is not searched for, the process proceeds to step S48 (step S46). When it is detected that the matching rule pattern is found, the distance calculating unit 3 determines that the classification target material belongs to the cluster of identification information corresponding to the matching rule, and classifies the classification target data into the cluster, and corresponds to the cluster database 5' The identification information of the clusters of the classified points is stored, and the classified object data classified is sorted (step S47). Further, when it is detected that there is no matching rule pattern, the distance calculating unit 3 detects the number of counts, that is, the identification information having the largest number of votes, and classifies the classified object data corresponding to the cluster of the identification information. Then, the distance from the five tenth parts is the cluster information database 5, and the classified object data is sorted corresponding to the identification information of the clusters of the home locations (step S 4 8 ). [Third Embodiment] The second embodiment is a table in which the distance from each cluster of the calculated classification target data is small (similarity is large) to the upper η -28-200818060. It is explained whether or not the clustering process of the respective classification target data is executed in accordance with the rule pattern present in the table. However, even if the feature amount set is set in each cluster as in the third embodiment, the operation corresponds to each feature amount set. The Mahalanobis distance is calculated as a correction distance, and a cluster of correction distances within a specific order of the upper rank is used as a cluster to which the classification target data belongs. In the following, the configuration of the third embodiment is the same as the first and second embodiments shown in Fig. 1, and the same reference numerals are given to the respective configurations. Only the operation different from the second embodiment will be described using the first embodiment. In the third embodiment, the processing of setting the above rule is not performed on the self-learning material, but the step S48 in Fig. 9 is directly executed. Fig. 10 is a flow chart showing an example of the operation of the cluster in the third embodiment. In the other cluster processing shown in FIG. 10, the processing from step S31 to step S34 is the same as the processing shown in Fig. 8, and the distance calculating unit 3, as described earlier, in step S34, In the order of small to large, the correction distance between the clusters in the internal memory unit is arranged, that is, the cluster information in which the correction distance of the classification target data is small is the upper order (step S34). Next, the distance calculating unit 3 detects the identification information of the cluster corresponding to each of the corrected distances from the small (upper) to the nth, and counts the number of pieces of identification information of each of the n included clusters, that is, for each cluster. The voting process is performed (step S55). Then, the distance calculating unit 3 detects the identification information of the maximum count 値 (the number of votes) in the voting result, and uses the cluster corresponding to the identification information, -29-200818060 as the cluster of the classified object data, and the cluster data. The library 5' corresponds to the identification information of the cluster of the home location, and the classified object data of the memory is memorized (step S56). Furthermore, the distance calculation unit 3 sets a threshold for the number of votes that the user has pre-screened for each piece of identification information. When the number of votes of the identification information with the highest number of votes is less than the threshold, even if it is executed, it does not belong to any The processing of clusters is also possible. For example, when the three clusters of clusters A, B, and C are classified, the number of votes relative to the identification information of cluster A is five, and the number of votes for the identification information of cluster B is three. When the number of votes with respect to the cluster C is two, the identification information of the maximum number of votes detects the cluster A and the distance calculating unit 3. However, when the above-mentioned threshold 设定6 is set for the cluster A, the distance calculation unit 3 is not satisfied with any cluster due to the number of votes of the identification information with respect to the cluster A being less than the critical number. Accordingly, in the cluster analysis of clusters in which the feature quantity is only slightly different from other clusters, the reliability of the classification processing can be improved for the cluster of the classified object data. [Method of Transforming Feature Quantity] Although it is expected that the parent group of each feature quantity performs a cluster analysis as a normal distribution, the presence or absence of a normal distribution 'the parent group has a skew distribution depending on the type (area, length, etc.) of the feature quantity, It is possible to calculate the distance between the classified object data and each cluster, that is, the accuracy of the classification object data and the similarity of each cluster -30-200818060. Therefore, depending on the feature quantity, some feature quantities need to be performed by a specific method to change the feature quantity of the parent group, so that the accuracy of the similarity is improved close to the normal distribution. As a transformation method for transforming to the normal distribution, an operation for calculating a feature quantity by calculating a η square root or a factorial or a number 値 by log or square root (/), cube root (3/), or the like Change in any of the formulas. Hereinafter, a method of converting each feature amount will be described using FIG. Fig. 1 is a flow chart showing an operation example of setting processing of each feature amount conversion method. Further, the conversion method is set for each cluster in units of feature numbers included in the cluster. Furthermore, the setting of the conversion method is performed using learning materials belonging to each cluster. The following processing will be described with the feature amount set cooperation unit 1 being executed, but other processing units corresponding to the processing may be provided. The feature amount set cooperation unit 1 uses the identification information of the cluster of classification objects as a key, reads the learning data contained in the cluster from the cluster database, and calculates (normalizes) the feature amount of each learning material (step S6 1 ) . Then, the feature amount set cooperation unit 1 performs the change of the feature amount by transporting the read learning materials by using any one of the arithmetic expressions for performing the feature amount conversion stored therein (step S62). When the conversion of the feature amounts of all the learning materials is completed, the feature amount set creating unit 1 calculates an evaluation 表示 indicating whether or not the distribution obtained by the conversion processing is close to the normal distribution (step S63). -31 - 200818060 Next, the feature quantity set cooperation unit 1 detects all the arithmetic expressions that are set in advance, that is, which are previously used as the conversion method, and whether or not the evaluation 算出 is calculated, and it is detected that the conversion characteristics are calculated in all the arithmetic expressions. When the evaluation of the acquired distribution is performed, the process proceeds to step S65, and when it is detected that all the calculation formulas have not yet been calculated, the processing of the next set expression is executed, so that the process is returned. Go to step S62 (step S64) ° When the feature amount conversion of all the arithmetic expressions is ended, the feature quantity set cooperation unit 1 detects that the distribution 値 is the smallest distribution among the distributions obtained in the set expression, that is, the most The conversion method for calculating the expression of the detected distribution is determined close to the distribution of the normal distribution, and the method for converting the feature amount of the cluster is internally set (step S65). The feature amount set cooperation unit 1 performs the above processing for each feature amount of each cluster, and sets a conversion method corresponding to each feature amount of each cluster. Next, the calculation of the evaluation 上述 in the above step S63 will be described using Fig. 12 . Fig. 1 is a flowchart showing an example of the operation of the process of evaluating the distribution of the distribution obtained by the arithmetic expression. The feature amount set cooperation unit 1 converts the feature amounts of the respective learning materials belonging to the object cluster by the set arithmetic expression (step S7 1 ). After the feature quantity of all the learning materials is converted, the feature quantity set cooperation unit 1 calculates the average 値μ and the standard deviation σ of the distribution (parent group) obtained by the converted feature quantity (step S72). Then, the feature quantity set cooperation unit 1 calculates the ζ値(1) from (x-μ) / σ using the average 値μ and the standard deviation σ of the above parent group (step -32 - 200818060 S73) 〇 then the 'feature quantity set The preparation unit 1 calculates the cumulative accuracy rate in the parent group (step S74). After the calculation, the feature quantity set cooperation unit 1 calculates the inverse function of the cumulative distribution function of the standard normal distribution by the cumulative accuracy in the parent group as ζ値(2) (step S75). Then the 'feature quantity set cooperation part 1 is to find the two ζ値 of the distribution of the feature quantity, that is, the difference between ζ値(1) and ζ値(2), that is, the error of the two ζ値 in the distribution (step S76). When the error of ζ値 is obtained, the feature quantity set cooperation unit 1 calculates the sum of the errors of the above two ζ値, that is, the sum of the errors (self-multiplication) as the evaluation 値 (step S77). The smaller the error of the above two flaws, the closer the distribution is to the normal distribution. If there is no flaw, the normal distribution. In addition, the more the distribution is out of the regular distribution, the larger the error. Next, before performing the cluster processing in the first to third embodiments, the feature amount of the classification target data will be described using FIG. Fig. 1 is a flow chart showing an example of the operation of calculating the feature quantity data of the classification target data. The distance calculating unit 3 extracts the feature amount of the recognition target from the input classification target data in accordance with the feature amount set set for each cluster, and performs all the normalization processing (step S 8 1 ). Next, the distance calculation unit 3 converts the feature amount used for classification into the cluster of the classification target by the conversion method (calculation formula) set for the feature amount of the cluster (step S82). - 200818060 Then, the distance calculation unit 3 calculates the distance of the cluster to be classified as described in the first to third embodiments (step S83). Next, the distance calculation unit 3 converts the feature quantity to the cluster of the classification target by the conversion method set corresponding to the feature quantity of each cluster, and performs the detection of whether or not to calculate the distance of the cluster by the feature quantity of the transformation. When the distance is obtained for all the clusters of the classification target, the process proceeds to step S85, and when the cluster in which the classification object remains is detected, the process returns to step S82 (step S84). Then, in each of the modes of the brothers 1 to 3, the processing from the point at which the calculation of the distance ends is started (step S 8 5 ). For the above reason, when the distance between the classification target data and each cluster is obtained in the Mahalanobis distance used in the present embodiment, since the expected feature amount is a normal distribution, each of the parent groups The closer the distribution of the feature quantities is to the normal distribution, the better the distance (similarity) between the clusters can be obtained, and the accuracy of classifying the clusters can be expected. EXAMPLES [Calculation Example] Next, using the cluster system of the first, second, and third embodiments, the classification accuracy of the conventional example was confirmed by the sample data shown in Fig. 14. It can be seen that although the number of samples is small, the accuracy of the conventional example or the above is obtained regardless of the amount of features used. In the first picture, 10 learning materials are defined as the types 1, 2, and 3 of each cluster, and each learning material has the feature quantities a, b, c, d, e, f, g, 8 of h. -34- 200818060 In this example, the learning material belonging to each cluster shown in Fig. 14 determines the feature quantity set used by the cluster, and then the learning set is also used as the classification object data to perform cluster analysis. The calculation result is that the first calculation method is the previous calculation method, and the characteristic quantities a and g which are the combination of the feature quantities are used, and the learning materials shown in the 14th picture of the cluster 1 to the cluster 3 are operated, and the Mahalano ratio is calculated. The distance is expressed as the judgment result. In Figure 15 (a), Cluster is listed as the Mahalanobis distance from Cluster 1, Cluster 2 is listed as the Mahalanobis distance from Cluster 2, and Cluster 3 is listed as the Horse with Cluster 3. Harlanobis distance. Further, the 'category list actually indicates the cluster to which each learning material belongs, and the judgment result indicates that the learning material and the Mahalanobis distance are the smallest clusters. The number indicating the correct classification type and the judgment result is the same characteristic quantity data. In Fig. 15(b), the number of the column indicates the cluster to which the learning material timepiece belongs, and indicates the number of the decision line. For example, "8" of the flag R1 is determined to be the cluster 1 of the clusters of 1 cluster, and the "2" of the marker R2 is the cluster 2 of the 10 clusters of the cluster 1. P0 is a ratio indicating the agreement between the positive solution and the answer, pl is the probability of the coincidence of the two, and k is the total correction rate, which is obtained by the following equation. Indicates that the higher the k, the higher the classification accuracy. k = (p0-p 1)/(1 -p 1) p0 = (a + d)/(a + b + c + d) pl = [(a + b) · (a + c) · (b + d) · (c + d)] · (a + b + c + d) 2 The relationship between a, b, c, and d in the above formula will be described using Fig. 16. The data belonging to cluster 1 is classified as cluster 1 and the number is a, and belongs to -35- 200818060. The data in cluster 1 is treated as cluster 2 and the number is b' a + b is the number of data belonging to cluster 1. Furthermore, the data belonging to cluster 2 is treated as cluster 2, the number is d, and the data belonging to cluster 2 is treated as cluster 1 and the number is c, indicating that c + d belongs to cluster 2. a + c is the number of clusters 1 that are classified into the full data a + b + c + d ' b + d is classified into the number of clusters b in the full data a + b + c + d. Next, in the seventh embodiment, using the calculation method of the first embodiment, the Mahalanobis distance is calculated for each learning material shown in Fig. 14 of the cluster 1 to the cluster 3, and the determination result is shown. The drawings (a) and (b) of the seventh embodiment are the same as those of the fifth embodiment, and thus the description thereof is omitted. It can be seen that the correct rate p0, the coincidence accuracy rate P 1 , and the total correction rate k are equal to the calculation method of the fifth figure. Here, a method of selecting a combination of the largest discrimination criterion 値λ for each cluster from the above-described combination is used, and a feature amount set corresponding to each cluster is calculated. The feature quantity set corresponding to the cluster 1 is a combination of the feature quantities a and h, and the feature quantity set corresponding to the cluster 2 is the feature quantity a and d, and the feature quantity set corresponding to the cluster 3 is A combination of feature quantities a, g is used. Next, in the eighth embodiment, using the calculation method of the second embodiment, the Mahalanobis distance is calculated for each learning material shown in Fig. 14 of the cluster 1 to the cluster 3, and the determination result is shown. The method for viewing the first (8) and (b) of FIG. 8 is the same as that of Fig. 15, and therefore the description thereof is omitted. The correct rate p 〇 is 0. 8 3 3 3, the probability of coincidence is p 1 is 〇 · 3 3 3 3, and the total correction rate k is 0. 75, it can be seen that the classification accuracy is improved compared to the conventional calculation method of Fig. 15. Here, from the combination of the above, the method of determining the feature amount for each cluster is selected by the method of determining the discrimination criterion 値λ to the upper third number in each -36-200818060 cluster. Three combinations of feature quantities a, h, a, g, d, and e are used as the feature quantity set corresponding to cluster 1, and three combinations of feature quantities a, f, a, d, a, b are used. As the feature quantity set corresponding to the cluster 2, three combinations of the feature quantities e, g, a, c, a, and g are used as the feature quantities corresponding to the cluster 3. Furthermore, the determination of the vote is arranged in a sequence in which the distance of the Mahalanobis is small, and the number of clusters entering from the few to the third is calculated, and the cluster of the largest number is regarded as the cluster to which the classification object data belongs. Next, in the nineteenth embodiment, using the calculation method of the second embodiment, the Mahalanobis distance is calculated for each learning material shown in Fig. 14 of the cluster 1 to the cluster 3, and the Mahalano is calculated. After the Bisz distance is multiplied by the correction coefficient (λ)_1/2, the rank of the distance is arranged to indicate the result of the determination. The method for viewing the seventeenth (a) and (b) of the present invention is the same as that of the fifteenth figure, and therefore the description is omitted. The correct rate p〇 is 0. 8 3 3 3, chance coincides that the pi is 0. 3 3 3 3, the total correction rate k is 0·75, and it can be seen that the classification accuracy is improved compared with the conventional calculation method of Fig. 15. Here, the method of selecting the feature quantity set for each cluster is selected by the method of selecting the discrimination criterion 値λ from the upper third number in each cluster. Three combinations of the feature quantities a, h, a, g, d, and e are used as the feature quantity set corresponding to the cluster 1, and three combinations of the feature quantities a, f, a, d, a, and b are used. As a set of feature quantities corresponding to cluster 2, three combinations of feature quantities e, g, a, c, a, g are used as feature quantities corresponding to cluster 3 - 37 - 200818060 Arranged in a sequence with a small distance from the Mahalanobis, the number of clusters entering from the few to the third is calculated, and the cluster of the largest number is regarded as the cluster to which the classification object data belongs. According to the results of the classifications shown in the above-mentioned fifth, seventh, eighth, and ninth aspects, it is judged that the present embodiment performs high-speed and high-precision cluster processing as compared with the conventional example, and the present embodiment is confirmed to be superior to the conventional example. Sex. [Application Example of the Present Invention] A. Inspection device Description The inspection device (for example, the defect inspection device) for classifying the inspection object, for example, the scratch type on the surface of the glass substrate, as shown in Fig. 20 is shown. Fig. 2 is a flowchart for explaining an operation example of selection of a feature amount set, and Fig. 2 is a flowchart for explaining an operation example in cluster processing. First, the selection action of the feature amount set will be described. The collection of learning materials in step S1 in the flowchart of Fig. 5 is a step S1 0 1 to step S1 0 5 corresponding to the flowchart of Fig. 21. Since steps S2 to S4 of Fig. 21 are the same as those of the flowchart of Fig. 5, description thereof will be omitted. By the operation of the operator, samples for learning materials corresponding to the clusters of the types of scratches to be classified are collected (step S 1 0 1 ) °. The illumination device 1 照射 2 illuminates the image acquisition unit 1 0 1 as learning materials. The shape of the scratches collected is obtained by the photographing device 103 from the image data of the scratched portion (step S102). Then, the feature data acquired by the self-image acquisition unit 1 0 1 calculates the feature amount of the scratch of the learning material (step S丨03}. The feature quantities of the acquired learning materials are classified into visuals. The obtained classification site 'executes the specifics of the learning materials in each cluster (step S 1 04 ) ° Then, the learning data of each cluster becomes a specific number (the number of samples set in advance) 'for example, each is about 30, and the steps are repeated. The process from si〇i to step S1 0 2 'when the number is a specific number, the cluster unit 丨〇2 executes all the processes after step S2 described in the fifth figure. Here, the clustering unit 1 〇5 is the first 1 or the cluster system in the second embodiment. Next, referring to Fig. 22, the cluster processing in the inspection apparatus of Fig. 4 will be described. Here, step S31 to step S34, S55, and S56 and Fig. 22 of Fig. 22 are described. The flowchart of Fig. 10 is the same, and the description is omitted. In the inspection apparatus of Fig. 20, when the inspection is started, illumination is performed on the illumination device 1 〇 2 for the glass substrate of the inspection object 100, and the photographing apparatus 1 〇 3 Photographing the surface of the glass substrate and taking the photo The image is output to the image acquisition unit 1 〇 1. In this manner, the defect candidate detection unit 104 detects a portion different from the plane shape in the photographic image input by the self-image acquisition unit, and sets the defect candidate to be classified ( Step S201) Next, the curve candidate detecting unit 104 cuts the image data of the candidate candidate portion from the captured image as the classification target data. Then, the defect candidate detecting unit 104 is the image data of the self-classifying target data. The feature amount is calculated, and the clustering unit 105 outputs the classification target data composed of the extracted feature amount sets (step S2 02). The subsequent cluster processing is explained in the step of the first map -39-200818060 Therefore, as described above, the inspection apparatus of the present invention can classify the scratches existing on the glass substrate into each type B of the scratches. Defect type judging device The defect type judging device shown in Fig. 23 is clustered in the cluster system of the present invention which has been described. The image capturing device 201 is composed of the illuminating device 102 and the photographic device 103 in Fig. 20 . The learned data of the place where the classification target data is classified is obtained, and is prepared in the cluster database of the cluster device 105. The feature amount set in Fig. 5 is also ended. The imaging device of the image capturing device mounted on each manufacturing device detects the defect candidate, and extracts the image data amount and outputs it to the data collecting device 203. After the classification data of the data collection device 203 of the control device 200 is transferred to the cluster, as described above, the cluster portion 105 is the classification target data input corresponding to the scratch cluster classification. C. Manufacturing management device The manufacturing management device of the present invention is set as shown in Fig. 24, manufacturing device 3 () 1, 3 〇 2, notification unit 303, poor condition device determining unit 305, and defect type determining device. . Here, the curve type determining means 3 〇 6 is the same as the above-described B item type determining means. Classes are made with high precision. The shape 105 corresponds to: the part 2 0 1 , the learning of each cluster 5 . Therefore, 202 is input and the extracted feature is input to the portion 1 〇 5. In the case of the type, the defect type determining device 306 described in the control recording unit 3 04, 306 is configured as the defect candidate determining device 306 in the corresponding defect candidate 104, and is provided in each of the manufacturing devices 3〇1. The photographing images of the image capturing devices 201 and 202 of the manufacturing apparatus extract the feature amount from the image and perform classification of the classified object data. Next, the poor device determining unit 305 has identification information indicating the diversity, and identification information indicating a cluster of points input from the defect type determining means 306 from the table in accordance with the relationship of the cluster occurrence factors. The cause of the occurrence is determined to be the device to be manufactured. In other words, the poor condition device determining unit 305 is a cause for detecting a defect in the manufacturing process of the product corresponding to the identification information, and then the device undone device determining unit 306 is operated by the notifying unit 303 and is in the recording unit. 3 04 corresponds to the date of the determination, the identification number of the defect classification, the cause of the occurrence, and the manufacturing identification information as the history. Further, the control device 300 is a manufacturing device that stops the determination by the good device determining unit 305, and controls the control parameters D. Manufacturing Management Apparatus The other manufacturing management apparatus of the present invention is constituted by the control unit 300, the manufacturing apparatuses 301 and 302, the notifying sections 303, 308, and the clustering unit 105 as shown in Fig. 25. Here, the clustering unit 1〇5 is the same as the configuration described in the :B term. The clustering unit 105 is different from the above-described cases of A to C, and the characteristic data of the image data is made by a manufacturing process of, for example, a glass substrate (material component, processing temperature, pressure, processing speed detecting unit 3 02, The table of the cluster, the classification of the classification is not the same as the state of the memory device, and the feature quantity of the above-mentioned A classification system, such as the recording system, is -41 - 200818060, according to the process The manufacturing state of each project describes that the feature amount is input to the clustering unit 1 1 5 by the engineering information set in each manufacturing device 301 or 302, and the clustering unit 105 is the item by the above classification. The glass manufacturing process in each of the manufacturing facilities is classified into a "normal state", "a defect that needs to be adjusted, and a state in which a danger is required to be adjusted". Then, 105, the notification unit 303 notifies the classification result to the identification information of the cluster of the classification result, and outputs the identification information of the clustering result to the control device, so that when the recording unit 309 corresponds to the date determined, the classification of the manufacturing state of each project is memorized. Identification number; the most versatile manufacturing condition; and the identification information of the manufacturing device. The control device 300 has a table indicating an adjustment item for returning the identification information of the cluster to the normal and an object of the data, and the cluster identification information input by the read cluster unit 105, the adjustment item of the manufacturing strip and the data, by the Read data control device. Further, even if the cluster system for realizing the recording in Fig. 1 is recorded on a computer-readable recording medium, the computer is attached to the recording medium, and the sorting and pair processing can be performed by performing the processing. Further, the term "computer system" as used herein includes hardware such as a machine. Furthermore, the "computer system" also includes a WWW system with an environment (or display environment). Further, the recording medium to be read refers to a floppy disk, a magnetic disk, a ROM, and a classification. The characteristic quantity of the sensor on it. After the feature quantity of the material is manufactured, the state of the whole state, the clustering component, and 300, and the characteristics and manufacturing conditions of the problem above the resume are corresponding to the function of the manufacturing system that returns to the normal correspondence. In the cluster OS of the recorded data or the surrounding area, there is a web page, "Computer can be CD-ROM, etc. -42-200818060, removable media, hard disk built in computer system, etc." The recording medium is a volatile memory (RAM) inside the computer system of the server or the client when the program is transmitted via a communication line such as a network or a telephone line via the Internet, and the program is kept for a certain period of time. Furthermore, the above program can be transmitted to another computer system via a transmission medium or a transmission wave in the transmission medium even if it is stored in a computer system such as a memory device. Here, the "transmission medium" of the transmission program is a medium having a function of transmitting information, such as a network (communication network) such as the Internet or a communication line (communication line) such as a telephone line. Furthermore, the above procedure may be used to implement one of the above functions. Further, even if it is possible to realize the above function by a combination with a program already recorded in a computer system, it may be a differential file (differential program). [Industrial Applicability] The present invention is applicable to high-precision classification and discrimination of information having a plurality of types of features, such as detection of defects such as glass articles, and can also be utilized for manufacturing state detecting devices or product manufacturing management. Device. [Brief Description of the Drawings] Fig. 1 is a block diagram showing a configuration example of a cluster system according to the first and second embodiments of the present invention. Fig. 2 is a table for explaining processing for selecting a feature set by discriminating the reference 値r. -43- 200818060 Fig. 3 is a table for explaining the process of selecting a feature set by discriminating the reference 値r. Fig. 4 is a histogram illustrating the effect of selecting a feature set by discriminating the reference 値r. Fig. 5 is a flowchart showing an operation example of a process of selecting a feature amount set for each cluster by the first embodiment. Fig. 6 is a flowchart showing an example of an operation of performing cluster processing on the classification target data by the first embodiment. Fig. 7 is a view showing an operation example of the table for generating the rule pattern used in the clustering processing in the second embodiment. Fig. 8 is a view showing an operation example of performing cluster processing on the classification target data by the second embodiment. Fig. 9 is a view showing an operation example of performing other cluster processing on the classification target data by the second embodiment. Fig. 10 is a flowchart showing an operation example of performing cluster processing on the classification target data by the third embodiment. Fig. 11 is a flow chart showing an example of the operation of the arithmetic expression for setting the method of converting the feature amount. Fig. 1 is a flow chart showing an example of the operation of calculating the evaluation 流程图 in the flowchart of Fig. 1 1 . The first 130 is a flowchart showing an example of an operation for calculating the distance of the feature amount converted by the set conversion method. Figure 14 is a table showing learning materials belonging to each cluster. Fig. 15 is a table showing the results of classifying the learning materials of Fig. 14 by the method of the conventional example -44-200818060. Fig. 16 is a view showing a method of calculating the total correction determination rate. Fig. 17 is a result table showing the results of the learning data of Fig. 14 by the cluster system in the first embodiment. Fig. 18 is a table showing the results of the learning materials classified by the cluster system in the second embodiment. Fig. 19 is a table showing the results of the learning materials classified by the clustering system in the second embodiment. Fig. 20 is a block diagram showing an example of the configuration of an inspection apparatus using the cluster system of the present invention. Fig. 21 is a flow chart showing an example of the selection operation of the feature amount set in the inspection apparatus of Fig. 20. Fig. 22 is a flow chart showing an example of the operation of the clustering processing in the inspection apparatus of Fig. 20. Figure 23 is a block diagram showing the constitution of a defect type determining device using the cluster system of the present invention. Fig. 24 is a block diagram showing a configuration example of a manufacturing management apparatus using the cluster system of the present invention. Fig. 25 is a block diagram showing an example of the configuration of another manufacturing management apparatus using the cluster system of the present invention. [Description of main component symbols] 1 : Feature amount set cooperation unit -45- 200818060 2 : Feature amount extracting unit 3: Distance calculating unit 4: Feature amount set memory unit 5: Cluster database 1 〇〇: Inspected object 1 〇1 : Image acquisition unit 102 : illumination device 1 0 3 . Imaging device 104: defect candidate detecting unit 105: clustering unit 2 0 0, 3 0 0 : control device 201, 202: image obtaining device 301, 3 02 : manufacturing device 3 0 3 : notifying unit 3 04 : recording unit -46-

Claims (1)

200818060 X. Patent application scope 1 · A cluster system belongs to a cluster system that classifies input data into various clusters formed by a parent group of learning materials by using the feature quantity of the input data, and has the following features: The feature amount set storage unit stores a feature amount set in which a combination of feature amounts used for classification is stored in the respective clusters; and the feature amount extracting unit extracts a feature amount 5 distance calculation unit set in advance from the input data, and corresponds to For each feature quantity set of each cluster, according to the feature quantity contained in the feature quantity set, the distance between the center of the parent group of each cluster and the input data is calculated to be output as each set distance; and the order is extracted In the ministry, the distances of the above collections are arranged from small to large. The cluster system according to the first aspect of the invention, wherein the feature amount set is set in a plurality of clusters. 3. The clustering system of claim 2, further comprising a cluster classification unit that classifies the input data by setting the order of the set distances in the set distances obtained for each feature amount set. The rule pattern of the classification criteria of each cluster is used to detect which cluster the input data belongs to. 4. The clustering system according to claim 3, wherein the cluster classification unit detects, by the rank of the set distance, which cluster the input data belongs to, and sets the rank as a set distance of the upper-47 - 200818060 A cluster system of the above-mentioned input data is detected as a cluster of a plurality of clusters. The clustering system described in claim 4, wherein the cluster classification section has a higher number relative to the rank. When the cluster that becomes the upper level is above the critical threshold, it is detected as the cluster to which the input data belongs. The clustering system according to any one of claims 1 to 5, wherein the distance calculating unit multiplies the correction coefficient set corresponding to the feature amount with respect to the collective distance, so that each feature The set distance between the sets of quantities is standardized. 7. The cluster system according to any one of claims 1 to 5, further comprising a feature quantity set cooperation unit for creating a feature quantity set of each cluster, wherein the feature quantity set cooperation part is for each For each of the majority of the combinations of the feature quantities, the average of the learning materials of the parent group of each cluster is taken as the origin, and the average distance between the origins and the learning materials of the parent group of the other clusters is obtained, and The combination of the feature quantities that will become the maximum average is selected as the set of feature quantities used to distinguish each cluster from other clusters. 8. A device for determining a defect type, characterized in that: the cluster system according to any one of the above-mentioned claims, wherein the input data is an image of a defect of the product 'by indicating a defect The feature quantity, the defect in the image data is classified according to the type of the defect-48-200818060 9 - The defect type determination device described in the eighth aspect of the patent application, wherein the above product is a glass article, and the defect of the glass article Classified according to the type of defect. I 〇 — 缺陷 缺陷 缺陷 缺陷 缺陷 缺陷 缺陷 缺陷 缺陷 缺陷 缺陷 缺陷 缺陷 缺陷 缺陷 缺陷 缺陷 缺陷 缺陷 缺陷 缺陷 缺陷 缺陷 缺陷 缺陷 缺陷 缺陷 缺陷 缺陷 缺陷 缺陷 缺陷 缺陷 。 。 。 。 。 。 。 II. A manufacturing state judging device characterized in that a defect type judging device as set forth in claim 8 or 9 is provided, and classification of defects of the product is performed, according to correspondence with an occurrence factor corresponding to the category, The cause of the defect in the process is detected. A manufacturing state judging device characterized by being provided with a cluster system as described in any one of claims 1 to 7, wherein the input data is a feature amount indicating a manufacturing condition in a process of the product. The feature quantity is classified according to the manufacturing state of each process of the process. The manufacturing state judging device according to claim 12, wherein the product is a glass article, and the feature amount in the process of the glass article is classified according to the manufacturing state of each process of the process. 1 . A manufacturing state detecting device characterized in that the manufacturing state judging device described in the first or second aspect of the patent application is provided, and the type of manufacturing state in each of the processes of the product process is detected. A product manufacturing management device is characterized in that: a manufacturing state judging device as described in the Patent Application No. 12 or 13 is provided, and a category of a manufacturing state in each of the processes for detecting a product process is performed. Corresponding to the control items of this category, process control in process engineering. -49-