JP4971811B2 - Function database generation method and function database generation device - Google Patents

Description

  The present invention relates to a function database generation method and a function database generation apparatus used for cluster analysis.

  Conventionally, there is a function cluster analysis as a method for classifying function curves based on the uneven shape of the curve, an increase or decrease tendency, and the like. In this function cluster analysis, a classification pattern of an arbitrary number of clusters can be derived from a rough classification with a small number of clusters to a fine classification with a large number of clusters.

In this method, function curves are classified according to the following procedure.
1). Enlarge, reduce, or translate the function curve to convert the function expression to a function curve that exists in the reference range or position 2). The distance between two types of functions ・ Integral value between functions ・ Defined based on Euclidean distance between the regression coefficients of the function expression, etc., and calculates the distance and matrix by all combinations of the functions to be classified, and calculates the distance matrix To do.
3). Using cluster analysis, the distance matrix is analyzed and classified into an arbitrary number of clusters.

A classification method using cluster analysis (for example, Patent Document 1) is provided for classifying pattern sets.
JP-A-8-194421

  By the way, in the above-described function cluster analysis, after performing the curve classification, the examiner compares the classification patterns, and the classification is not too rough and not too fine, or there is a curve having a similar shape that is characteristic of each cluster. Whether or not the determination of the number of clusters becomes ambiguous, and different examiners have different numbers of clusters to be selected. The method disclosed in Patent Document 1 does not perform curve functioning.

  The present invention has been made in view of the above points. The purpose of the present invention is to provide a function database in which the optimal classification number is quantitatively determined using a numerical index when a function curve is classified into various clusters. It is an object of the present invention to provide a function database generation method and a function database generation apparatus that can be generated.

In order to achieve the above-described object, the invention according to the function database generation method of claim 1 is based on the function cluster analysis results of a plurality of function curves, and the function cluster analysis results are used as an index indicating the degree of grouping for each cluster of the function cluster analysis results . The distance sum is calculated, and the number of clusters when the convergence rate of the intra-cluster distance sum when the number of clusters increases by 1 exceeds a predetermined value is determined as the number of classes, and the function curve is determined with the determined number of classes. It is characterized by creating a database.

According to the invention of the function database generating method according to claim 1, function database function curve upon classified into various clusters, as the optimum number of classifications that may be quantitatively determined to using numerical indicators classification number Can be generated. In addition, since the intra-cluster distance sum is used as an index, the degree of unity can be evaluated with a simple calculation.

In the invention according to the function database generation device of claim 2, a function cluster result acquisition unit that acquires a function cluster analysis result of a plurality of function curves, and an intra- cluster distance as an index indicating the degree of grouping for each cluster of the function cluster analysis result A total number is calculated, and when the number of clusters increases by 1, the number of clusters when the convergence rate of the intra-cluster distance sum exceeds a predetermined value is determined as the number of classifications, and the determined number of classifications And a database output unit that outputs the function curve in a database.

According to the invention of the function database generation apparatus according to claim 2, function database function curve upon classified into various clusters, as the optimum number of classifications that may be quantitatively determined to using numerical indicators classification number Can be generated automatically.

In addition, since the intra-cluster distance sum is used as an index, the degree of unity can be evaluated with a simple calculation.

In the invention according to the function database generation apparatus according to claim 3, characterized in that in the invention of claim 2, a rear shape input unit that enter multiple function curve representing the human back shape function, the function cluster analysis for the plurality of function curve And a function cluster analysis unit for performing the above.

According to the function database generating apparatus of the third aspect , it is possible to generate a function database for classifying the back surface shape by simply inputting a plurality of back surface data.

According to the function database generating method invention, when the classifying function curve to various clusters, it is possible to generate a function database as the best classification number can be determined quantitatively using numerical indicators classification number The invention relating to the function database generation device has an effect that the function database can be automatically generated.

Hereinafter, the present invention will be described with reference to an embodiment of a function database generation device.
(Embodiment 1)
The function database generation device A of the present embodiment is for use in, for example, classification of the back shape of the human body. As shown in FIG. 1, the back surface shape curve input unit 1, the function curve input unit 2, and the arithmetic processing unit 3 and a result display unit 4.

  As shown in FIGS. 2A and 2B, the back surface shape curve input unit 1 is a back surface shape measuring device 100 that measures the displacement of each probe 101 that presses the back surface of the measurement subject M against the back surface, or the back surface. A back surface shape data source 10 composed of a shape database (not shown) and a plurality of back surfaces that approximate a plurality of back surface shape data acquired from the back surface shape data source 10 to generate a plurality of back surface function curve data D1. And a shape curve creating unit 11.

Function curve input section 2 is for inputting a plurality of function curve data D2.

  The arithmetic processing unit 3 includes a function cluster analysis unit 30, a function cluster result acquisition unit 31, a classification number determination unit 32, an output graph creation unit 33, and a database creation unit 34.

  The function cluster analysis unit 30 classifies the back surface shape function curve data D1 from the back surface shape curve input unit 1 or the function curve data D2 from the function curve input unit 2 into various cluster numbers by function cluster analysis.

  The function cluster result acquisition unit 31 obtains basic information such as the result of the number of clusters classified by the function cluster analysis unit 30, the function expression of the function to be examined (for example, regression coefficient), the examination interval (a1 ≦ x ≦ a2), and the like. get.

  The classification number determination unit 32 calculates an index of the degree of unity when classified into each cluster number, and determines the optimum classification number from the change of this index with respect to the number of clusters.

  The output graph creation unit 33 creates a graph or a table showing the number of clusters calculated by the classification number determination unit 32 and the change in the unity degree index.

  The database creation unit 34 creates a function database by converting the function curves studied with the optimum number of classifications obtained by the classification number determination unit 32 into a database.

  The result display unit 4 displays / outputs the optimum classification number obtained by the classification number determination unit 32 of the arithmetic processing unit 3, the graph created by the output graph creation unit 33, and Table 1 described later.

Next, in the function database generation apparatus A of the present embodiment, the processing from obtaining the result of cluster analysis of a plurality of back surface shape function curve data to displaying the result such as the optimum classification number is shown in the flowchart of FIG. This will be explained based on.

  First, in step S1 of inputting a plurality of function curve data and function cluster analysis results, the classification number determination unit 32 determines the function formula (regression coefficient, etc.) of the function to be examined, information such as the examination interval a1 ≦ x ≦ a2, and the function cluster. Get the results for each number of clusters in the analysis.

  In the next step S2 for calculating the grouping degree index for each number of clusters and the step S3 for determining the optimal number of clusters, the classification number determination unit 32 determines the number of clusters for which the change in the index relative to the change in the number of clusters is equal to or less than the determination reference value. Is determined as the optimum classification number. As a result, the optimum classification number can be automatically determined. It is assumed that the determination criterion is set in the arithmetic processing unit 1 in advance.

  Here, as an index of the unity degree, the intra-cluster distance sum can be used as shown in FIG. The intra-cluster distance sum is a sum of distance sums of all combinations of two functions belonging to the same cluster in all clusters. By using this intra-cluster distance summation, it is possible to evaluate the degree of grouping of functions with a simple calculation. The numbers in the circles in FIG. 4 indicate cluster numbers, and in the illustrated example, four clusters are indicated.

  Now, the degree of unity can be evaluated using the non-conformance statistic of the index of unity degree or the non-conformity statistic of the sum of distances within the cluster, which is one example of the index of unity degree. Here, the non-conformity statistic is uniform in the same space as the function group data for examining the optimum classification number and the actual function group data for examining the optimum classification number as shown in FIG. There is data shown in FIG. 5B in which the distance data of the function distributed in is created by simulation. Circles in the figure indicate each cluster.

  For each, a value representing the degree of grouping of the function curve is calculated for each number of clusters. For each number of clusters, the difference in values representing the unity degree calculated from the non-conformity statistics in FIGS. 5 (a) and 5 (b) is compared. As a result, the degree of classification between clusters can be increased, and a database that can be classified clearly and accurately can be generated.

  By repeating the simulation of non-conformance statistics shown in FIG. 5 (b) multiple times and comparing the average value and median of the non-conformance statistics at each time, the accuracy of the non-conformance statistics further increases. A more optimal number of clusters can be determined.

  Using the regression coefficient of the function curve as the distance to calculate the cluster distance sum, Euclidean distance (square root of difference, so-called n-dimensional linear distance), Euclidean distance square (sum of squared difference), urban area The sum of distances such as distance (sum of absolute values of differences) can be used.

The distance for calculating the total distance within the cluster can also be calculated by using an integrated value of the difference between two function curves for calculating the distance. In addition, as a distance for calculating the intra-cluster distance total, a section in which the function cluster analysis is performed with two functions y = f _i (x) and y = f _j (x) (a ≦ x ≦ b) for calculating the distance [ a, b] in x ₀ <x ₁ <x ₂ <... <x _n {f _i (x ₁ ), f _i (x ₂ ),..., f _i (x _n )} and {f _j (X ₁ ), f _j (x ₂ ),..., F _j (x _n )} distance (Euclidean distance, Euclidean distance square, city distance, etc.) can also be used as an integrated value (FIG. 6 shows an example thereof) Showing).

  The intra-cluster distance sum calculated in this way is compared for each number of clusters, and a change in value, for example, where the convergence rate of the Euclidean distance sum when the number of clusters increases by 1 exceeds 80% is determined as the optimal classification number. You can also.

Now, after determining the optimum number of classifications in step S 3 described above, in the next step S4, the output graph creation unit 33, in step S2, the number of each cluster calculated by the classification number determination unit 32, the coherent degree A graph showing the change of the index from the index (FIGS. 7A and 7B ) and a table as shown in Table 1 are created.

  When the graph creation step S4 is completed, the database creation unit 34 creates a database of function curves studied with the optimum number of classifications obtained by the classification number determination unit 32 (S5).

  When this database creation step S5 is completed, the result display unit 4 displays and outputs the optimum classification number obtained by the classification number determination unit 32 and the graph and table created by the output graph creation unit 33 (S6).

  As described above, the function database generation apparatus A according to the present embodiment can generate a function database in which the optimal classification number is quantitatively determined using a numerical index when the function curve is classified into various clusters. .

  By the way, the function database generated by the above-described function database generation apparatus A is used for the back surface shape classification determination apparatus 5 shown in FIG. 8, for example. Next, the back surface shape classification determination apparatus 5 will be described.

  As shown in the figure, the back surface shape classification determination device 5 is composed of a microcomputer or the like, and includes an arithmetic processing unit 50 that performs arithmetic processing and hardware control, a readable / writable storage unit 51, and a display unit capable of graphic display. 52 and communication for acquiring from the outside the back shape data of the determination target measured by the back shape measurement apparatus 100 existing on the network 6 such as the Internet, and the back shape data of the determination target stored in the database 7 Part 53. The communication unit 53 may not be provided when data acquisition from the network 6 is not performed. In FIG. 1, the interface between the arithmetic processing unit 50, the storage unit 51, the display unit 52, and the communication unit 53 is not shown.

The storage unit 51 stores, as basic data 51a, back surface shape data obtained by functionalizing the unevenness of the back surface of the human body necessary for performing the determination processing as a plurality of clusters, and the basic data 51a is stored in the present invention. constituted by a function database (back shape database) which thus generated function database generating apparatus a. In addition, numerical values that are criteria for determining the classification of the back surface shape, a matrix table for point conversion described later (see Table 4), and the like are also stored as basic data 51a.

  Furthermore, the storage unit 51 acquires and stores measurement data of the target back surface shape from the back surface shape measurement device 3 and the database 4 on the network 2 via the communication unit 53, or stores the measurement data in the back surface shape classification determination device 1. Measurement data of the back surface measurement data to be judged input using the input means (not shown) and data such as the start point (top) / end point (bottom end) that define the range to be analyzed 11b is stored.

  The calculation processing unit 50 has a distance calculation function 50a that calculates the distance between the measurement data to be determined and the back surface shape data that is a determination criterion divided into a plurality of clusters as a function for the classification determination calculation process, Based on the distance data calculated by the calculation function 50a, the determination processing function 50b for determining to which cluster the back surface shape data to be determined belongs, and the processing results of both the processing functions 50a and 50b, and the two-dimensional coordinates An output calculation function 50c that performs processing such as calculation of map coordinates and conversion of the number of posture points is provided.

  The display unit 52 displays a report based on the result determined by the determination processing function 50b and the calculation result calculated by the output calculation function 50c as described later.

  Next, the classification determination operation of the back surface shape classification determination device 5 will be described with reference to the flowchart shown in FIG.

  First, the arithmetic processing unit 50 acquires back surface measurement data to be determined from the back surface shape measuring apparatus 100 and the database 7 on the network 6 through the communication unit 53 or determines from the measurement data 51 b of the storage unit 51. The measurement data of the back surface shape is acquired (S1).

  Here, a pre-process S2 is performed in order to correct the height of the person to be measured and the difference in the standing position of the person to be measured with respect to the back shape measuring apparatus 100 used for measurement.

  In this pre-processing S2, only measurement data in the range from the start point to the lower end of the buttock is cut out from the measurement data based on the start point (top) / end point (lower end of the buttock) that defines the range to be analyzed acquired together with the measurement data.

  After the cutout, the cutout data is enlarged or reduced to a specified length (y-axis), and the horizontal direction (projection amount) is also enlarged or reduced to correct the height. For example, when height correction is performed with a length of 100, correction is made as y-axis = (probe number−start point) / (100 / (end point−start point)), protrusion amount / (100 / (end point−start point)).

  After this height correction, in order to correct the error before and after the standing position, the x-axis processing is performed so that the median value of the height-corrected data is brought to zero. In this case, correction is performed in the form of x axis = protrusion amount−median value of the protrusion amount.

  Now, processing for obtaining an approximate curve for obtaining a smooth curve is performed using the data that has undergone the pre-processing S2 (S3). In this process S3, an approximate expression is obtained using, for example, ninth-order function approximation, and a predicted value is obtained from this approximate expression. This predicted value is used as back surface measurement data to be determined in subsequent processing. It should be noted that automatic determination may be made by referring to the difference between the raw measurement data, the spline approximation curve, or the like as to what order function approximation is optimal for each person to be measured.

  Up to this point, the function of the arithmetic processing unit 50 as the back surface shape data acquisition unit of the arithmetic processing device 1 is performed.

  After completing this processing S3, the arithmetic processing unit 50 reads the basic data 51a from the storage unit 51 by the distance calculation function 50a, calculates the distance between the back surface shape data of each cluster and the back surface shape data to be determined, Using the calculation result, the determination processing function 50b determines to which classification (cluster) the back surface shape data to be determined belongs (S4).

  In this case, for example, the average value in the x-axis direction in the n back surface shape data belonging to each cluster from the basic data 51a of the storage unit 51, or the representative value of each cluster consisting of the median value, and the back surface shape data to be determined Distance (Euclidean distance, Euclidean square distance, city area distance, etc.) is calculated, and the classification (cluster) with the smallest distance is selected.

  After the selection of the cluster, the distance calculation function 50a performs a process of calculating a separation value between the determination target back surface shape and the ideal back surface shape (S5).

  In this process, the ideal back surface shape data registered as the basic data 51a [the representative value of the best cluster (for example, the shape of the approximate straight line is nearly vertical and has little unevenness) (n names belonging to the cluster) The distance between the back surface shape data to be judged and the back surface shape data of an individual determined as an ideal shape by a researcher or other expert regarding the posture) Euclidean distance, Euclidean square distance, city area distance, etc.) are calculated as the difference (separated value) from the ideal shape. By this calculation, the deviation state of the determination target back surface shape with respect to the ideal back surface shape is evaluated.

  After this calculation, the calculation processing unit 50 uses the output calculation function 50a to drop the back surface shape data into two-dimensional data, calculates the coordinates, and applies a rotation to correlate with the index that is characteristic of the x-axis. To calculate map coordinates (S6).

  In this case, the coordinates are based on the distance (Euclidean distance, Euclidean square distance, city area distance, etc.) between the representative values of each cluster (average value or median value in the x-axis direction in the n back surface shape data belonging to each cluster). Is calculated.

  Table 2 shows the Euclidean square distance between the medians of the seven classifications (clusters), and the calculation of the above-described coordinates will be described in detail based on Table 1. The numbers in “” in the table indicate the cluster numbers.

  Here, the positions Pi of the clusters “1” to “7” are set as follows.

P1: (x1, y1)
P2: (x2, y2)
P3: (x3, y3)
P4: (x4, y4)
P5: (x5, y5)
P6: (x6, y6)
P7: (x7, y7)
If the coordinates of cluster “2”, which is an ideal back surface cluster, is defined as (0, 0), the other point is cluster “1”, and the y coordinate is fixed to 0, the following simultaneous equations are obtained from the three-square theorem. The equation holds.

x3 ² + y3 ² = d23 (1)
x4 ² + y4 ² = d24 (2)
x5 ² + y5 ² = d25 (3)
x6 ² + y5 ² = d26 (4)
x7 ² + y7 ² = d23 (5)
(X3-√d12) ² + y3 ² = d13 (6)
(X4-√d12) ² + y4 ² = d14 (7)
(X5-√d12) ² + y5 ² = d15 (8)
(X6-√d12) ² + y6 ² = d16 (9)
(X7−√d12) ² + y7 ² = d17 (10)
The numbers after x and y indicate the cluster number, d indicates the distance between the two points, and the two numbers after the two indicate the corresponding cluster numbers of the two points.

  Since the distance d between the clusters is clear from Table 2, all the x and y are obtained by solving the simultaneous equations (1) to (10). In the above example, the following coordinates are obtained. It was.

P1: (−1.1964, 0)
P2: (0, 0)
P3: (1.2862, 0.76141)
P4: (2.1208, 2.6698)
P5: (−0.67218, 1.5764)
P6: (1.00087, 1.9164)
P7: (0.24599, 0.94683)
Next, the coefficient of the approximate curve in the back shape data of the median value of each cluster is obtained. Table 3 shows the obtained results. The coefficients shown in Table 3 are negative if the posture is backward tilted, positive if the posture is forward tilted, and close to 0 if the posture is close to straight. As a result, when the x-axis represents an inclination, it can be said that the clusters “2” and “5” are positioned so as to be substantially aligned on the y-axis.

  FIG. 8 shows a plot of the two-dimensional coordinates of the clusters “1” to “7”, where each black dot is the obtained coordinate, and the number given to each indicates a cluster number. In the figure, the solid curve represents the back surface of each cluster, and the broken line represents the ideal back surface curve.

Here, for example, as shown in FIG. 9, in order to bring the coordinates P5: (x5, y5) of the cluster “5” on the y axis where x = 0, the coordinates P2 should be rotated clockwise. The rotation angle θ can be expressed as θ = cos ⁻¹ (y5 / √d25), and by solving this, θ = 23.1 ° is obtained.

And the coordinates P5 ′ after rotation: (x5 ′, y5 ′) are
x5 ′ = x5 cos θ−y5 sin θ
y5 ′ = x5sin θ−y5 cos θ
It becomes.

  Then, new coordinates P1 'to P7' when the other coordinates P1 to P7 are rotated clockwise by 23.1 ° as described above with the coordinate P2 as the center are as follows.

P1 ′: (−1.1005, 0.46926)
P2 ': (0, 0)
P3 ′: (1.4818, 0.19593)
P4 ′: (2.998, 1.6242)
P5 ′: (0, 1.7138)
P6 ′: (1.6795, 1.3672)
P7 ′: (0.59764, 0.77448)
Similarly, the coordinates can be obtained by using the distance from the cluster “1” and the data to be determined from the cluster “2”.

  After the map coordinate calculation process (S6) is completed as described above, the calculation processing unit 50 calculates the number of posture points using the output calculation function 50c (S7).

  In this calculation process, first, the slope of the approximate curve of the back surface shape data to be determined is the x-axis, the departure value (distance) from the ideal obtained in S5 is the y-axis, and from the separation value, the basic data 51a is stored in the storage unit 51. The posture score is calculated using a matrix table (Table 4) showing the relationship between the stored distance and the posture score.

   It should be noted that points may be assigned by weighting each item such as the head, waist and buttocks.

  Thus, the arithmetic processing unit 50 informs the user through the display unit 12 based on the result of classification (cluster) determination by the determination processing function 50b and the calculation result of the two-dimensional coordinates and posture points by the output arithmetic function 50c. The report discharge process to be presented is performed (S8).

  The measured person who sees this report can intuitively understand the degree of separation between his / her ideal posture and the ideal posture.

It is a block diagram of one Embodiment of the function database production | generation apparatus of this invention. (A) is a side view of the back curve measuring apparatus used for one Embodiment, (b) is a side view of the use condition of a back curve measuring apparatus. It is a flowchart for operation | movement description of one Embodiment. It is explanatory drawing of the distance total in a cluster used for one Embodiment. It is explanatory drawing of the conformity degree statistics used for one Embodiment. It is an example figure of the distance calculation method used for one Embodiment. It is a result output example figure of one Embodiment. It is a block diagram of the back surface shape classification | category determination apparatus using the function database produced | generated by one Embodiment. It is a flowchart for operation | movement description of a back surface shape classification determination apparatus.

Explanation of symbols

A function database generator 1 back surface shape curve input unit 10 back surface shape data source 11 back surface shape curve creation unit 2 function curve input unit 3 arithmetic processing unit 30 function cluster analysis unit 31 function cluster result acquisition unit 32 classification number determination unit 33 for output Graph creation unit 34 Database creation unit 4 Result display unit

Claims (3)

From the function cluster analysis results of a plurality of function curves, the intra- cluster distance sum is calculated as an index indicating the degree of grouping for each cluster of the function cluster analysis results, and the convergence rate of the intra-cluster distance sum when the number of clusters is increased by one A function database generation method characterized in that the number of clusters when the value exceeds a predetermined value is determined as the number of classifications, and the function curve is databased with the determined number of classifications. A function cluster result acquisition unit for acquiring a function cluster analysis result of a plurality of function curves, and an intra- cluster distance sum as an index indicating the degree of grouping for each cluster of the function cluster analysis result, and when the number of clusters increases by one A number-of-classes determination unit that determines the number of clusters when the convergence rate of the intra-cluster distance sum exceeds a predetermined value as a number of classifications; and a database output unit that outputs the function curve in a database with the determined number of classifications; A function database generation device comprising: 3. The apparatus according to claim 2 , further comprising a back surface shape input unit that inputs a plurality of function curves representing the back surface shape of the human body as a function, and a function cluster analysis unit that performs a function cluster analysis on the plurality of function curves. function database generation equipment described.

Images