JPH0132464B2 - - Google Patents

Description

[Detailed Description of the Invention] [Summary] Development of a paper sheet validating machine that collects data on paper sheets including banknotes and performs various processes in developing a validating machine for the paper sheets. The system has a function to check the validity of the collected data and then create sample data, and can also use the sample data to perform a simulation that applies recognition logic corresponding to the identification machine to be tested. This disclosure discloses a configuration in which a discriminator to be tested can be tested and evaluated.

[Industrial application field]

The present invention relates to a paper sheet classification machine development system, particularly,
A paper that allows you to test and evaluate the identification machine to be tested by collecting data and creating desired sample data, and using the sample data to perform a simulation that activates the recognition logic corresponding to the identification machine to be tested. This is related to a leaf identification machine development system.

[Conventional technology]

Conventionally, recognition machines for paper sheets including banknotes have been developed. However, in the case of conventional technology, a large amount of data similar to that used by the classification machine under development is collected from the paper sheets to be classified, and the classification machine under development I was trying to find out how accurately I could identify paper sheets. In other words, by setting in advance a so-called preferable discrimination mode, and more specifically, by setting the sensor position in advance at a preferable position, it is possible to determine how accurately the discrimination mode in such a discrimination machine can perform discrimination. The method used was to find out whether

[Problem that the invention is trying to solve]

In the conventional case described above, it has not been possible to develop a discriminator of the desired form from a comprehensive standpoint. For example, in a paper identification machine that discriminates paper sheets, it is necessary to evaluate which position on the paper sheet is best for the sensor to sense.
This was extremely difficult and virtually impossible because no data was collected when the sensor position was changed.

[Means for solving problems]

The present invention is intended to solve the above points,
In particular, it allows the creation of sample data suitable for various simulations. That is, as will be described later, the surface of the paper sheet is T _j Z _k (j=1, ..., i,
After collecting data for each square area divided into square areas given by i+1,...;k=1,2,...) for a large number of sheets, The data about the square area of
A desired amount of variation is given as appropriate and generated as specimen data so that it can be used in developing a discriminator.

FIG. 1 shows a basic configuration diagram of the present invention. Reference numeral 10 in the figure is a sample data creation section, which has a function of creating sample data suitable for various simulations based on the collected data.

Reference numeral 101 is a sample extraction unit, which introduces a variation amount into the collected data in order to eliminate deviation due to the characteristics of the sensor at the time of collection, and based on the result, performs subsequent processing. Extract as many samples as are suitable for the process.

Reference numeral 102 denotes a sensor variation extraction unit, which takes into consideration and cancels the variation in the characteristics of the sensor in the test object discrimination machine, in a simulation corresponding to a standard test object discrimination machine, for example, when the classification rate is set to an arbitrary predetermined value. Introduce the amount of variation so that it can take any value.

Reference numeral 103 is a reference value creation unit that uses the standard sample obtained by the sensor fluctuation extraction unit to
For example, if you run a simulation that corresponds to the above-mentioned standard test target identification machine, the classification rate will be 50% or
In order to reach each value such as 80%, a standard value is created to determine what kind of discrimination conditions should be given.

Reference numeral 104 denotes a classification rate creation unit/misclassification rate creation unit, which instructs the test target classification machine to perform a classification rate simulation or a misclassification rate simulation.

Reference numeral 105 determines whether or not to modify the discrimination logic based on the result of the simulation.

Reference numeral 106 evaluates in which track and in which zone data continuity is lost on the paper sheet side, by utilizing the results of the above-mentioned simulation.

[Effect]

From the data collected and stored in the database, it is necessary to create specimens suitable for various simulations. To this end, there may be some kind of bias in the characteristics of the sensor of the data collector, and it is necessary to introduce a variation amount to eliminate this bias. Sampling section 1
01 introduces the amount of variation within the range of maximum +a% to minimum -b% for each track and/or each zone obtained for a large number of paper sheets, and the resulting average value M ′ and the variance V′ are obtained. Then, based on the data into which the amount of variation has been introduced, the data is reduced to a number suitable for subsequent simulations to obtain a sample, without changing the characteristics of the population of the data as much as possible.

Using the samples obtained in this way, samples suitable for various simulations are extracted. However, taking into consideration that there are variations in the sensors of the test object identification machines, it is necessary to use a standard test object identification machine for example. In order to obtain a high discrimination rate, a standard sample is created in which variations corresponding to all combinations of the upper and lower limits of the sensor's variation range are introduced. The sensor fluctuation extraction unit 102 performs processing for this purpose.

The reference value creation unit 103 determines, based on the standard sample, what kind of discrimination conditions should be given to the standard test object discrimination machine to obtain, for example, a desired discrimination rate.

〔Example〕

FIG. 2 shows the configuration of an embodiment of the paper sheet discriminator development system of the present invention.

In the figure, reference numeral 1 is a data processing device, 2 is a bill data collector, 3 is a data editing device, 4 is a floppy disk that holds edited information and supplies it to the data processing device 1, and 5 indicates a banknote validating machine that is being tested.

The banknote data collector 2 sets m×n grid areas on both the front and back sides of the banknote, and collects data for each grid area, such as color, amount of light reflection by color, amount of light transmission by color, etc. Collect. Furthermore, paper sheets are set so that data can be collected for each square area. The data editing device 3 adds additional data such as the denomination, setting direction, environmental conditions at the time of data collection, and banknote manufacturing lot number to the collected data, and writes the data onto the floppy disk 4.

The data processing device 1 is a floppy disk 4
It has the function of reading the contents of the data into the database 6 and performing the following processing. That is, 1) Conveyance state evaluation unit 7...When a banknote is set in the banknote data collector and data about the banknote is collected, whether or not the banknote conveyance speed is extremely abnormal. , check the degree of skew during conveyance.

2) Data base information evaluation unit 8... Regarding the information group collected on the database 6,
(i) Acquisition of input data, such as whether there are an extremely large number of banknotes of a particular denomination, or whether there is an extremely large amount of data collected under a certain setting condition when banknotes are set. (ii) checking for data anomalies, such as when a 10,000 yen bill was incorrectly set, but a 1,000 yen bill was incorrectly set during data collection; The reliability of the information population on the database 6 is evaluated by checking for undesired deviations and data abnormalities in the data.

3) Banknote side data change monitoring unit 9...Although it cannot be detected by the human eye, the type of ink on banknotes is sometimes changed. For this reason, it is necessary to constantly monitor the presence or absence of such changes and to ensure that the discrimination function of the discrimination machine correctly follows the changes. The sample data described later will be changed to preferable data.

4) Sample data creation unit 10: Creates a preferred number of sample data that can accommodate sensor variations in the discrimination machine to be tested from the collected information, and supplies the data to the recognition logic unit 12, which will be described later.

5) Discrimination data synthesis unit 11... When collecting data by the banknote data collector 2, it is extremely difficult to collect a large amount of data when, for example, a banknote is skewed at a predetermined angle. It's complicated. For this reason, it is desirable to synthesize data obtained when the vehicle is skewed at the predetermined angle from a group of data collected without skew.
In addition, there may be times when you want to perform a test such as changing the placement position of the sensor in the banknote validator, and it is practically impossible to change the sensor position of the data collector 2 and collect data again. Can not. For this purpose, the data when changing the sensor position is stored in the above data base 6.
It is desirable to synthesize the above data as appropriate.
The discrimination data synthesis section 11 has such a function.

6) Recognition logic section 12... When developing a banknote validator, it is necessary to test the degree of recognition function that the banknote validator has. This requires the data collected by the data collector 2 to be examined by applying a preferred form of epistemic logic. In addition, the sample data creation section 1
It is necessary to examine the data created in 0 by applying recognition logic. The recognition logic unit 12 has a function for this purpose.

7) Discrimination sensor evaluation unit 13... evaluates what kind of sensor is more preferable as a sensor to be used in the test object discrimination machine.

The data processing device 1 has the above-mentioned processing functions, but as shown in the figure, (i) the database internal information evaluation section 8 has the input data acquisition condition checking section 8-
1, data abnormality check section 8-2, etc.
(ii) The discrimination data synthesis unit 11 includes a data synthesis unit 11-1 for sensor position movement, a data synthesis unit 11-2 for skew movement, etc., and (iii) the recognition logic unit 12 performs discrimination processing based on distance. It has a processing section 12-1, a processing section 12-2 based on intra-group division, and the like.

The illustrated banknote validating machine 5 is a valid validating machine to be tested,
This corresponds to the discriminating machines that are actually in use and in operation, and information from these discriminating machines 5 is guided to the data processing device 1 via the data editing device 3 as necessary.

FIG. 3 shows the configuration of an embodiment of the banknote data collector, in which FIG. 3A is a side view, FIG. 3B is a plan view, and FIG.
Figure C shows a configuration diagram of the roller. Reference numeral 14 in the figure is a feeding section, which supplies the banknote 18 to the data collection section 15 when the banknote 18 is inserted. The data collection unit 15 includes a roller 19 as shown in FIGS. 3B and 3C.
is generated, and the banknote 18' is conveyed on the conveyance path 20 as shown by the arrow. The conveyance path 20 is shown in FIG. 3B.
As shown, it has a width that is sufficiently large compared to the distance in the longitudinal direction of the banknote 18'. On the conveyance path 20, banknote entry detection sensors S ₁ and S ₂ 21 and passage detection sensors 22 are provided, and data collection sensors 23 are arranged symmetrically with respect to the vertical center line shown in FIG. 3B. ing. A track detection sensor 24 is also provided as shown.

As described above, since the width of the conveyance path 20 is sufficiently large compared to the longitudinal length of the banknote 18', the upper end of the banknote 18' is at the upper end of the conveyance path 20 shown in FIG. 3B. The banknotes 18' can be placed in any desired position, from a state in which the banknotes 18' are set and transported so that they touch each other, to a state in which the banknotes 18' are set and transported in such a way that the lower end of the banknotes 18' touches the lower end of the transport path 20 shown in FIG. It is possible to set Corresponding to these set positions, the position on the banknote 18' that the data collection sensor 23 senses is determined.

As shown in FIG. 4, the top of the banknote 18' is divided into m×n grid areas 25, and the _grid areas arranged in the horizontal direction in _FIG . If the grid area groups arranged in the vertical direction in FIG. 4 are defined as zones Z ₁ , Z ₂ , Z ₃ . Thus, while the banknote 18' is being conveyed, the data collection sensor 23 collects data in the square areas on the two upper and lower tracks. And banknote 1
8' on the conveyance path 20, each of the square areas 25 within a predetermined range on the banknote 18' is
5 data can be collected. The data in the hatched square area 25 in FIG. 4 is extracted at the transport timing corresponding to zone Z ₂ while one of the sensors 23 is collecting data along the track T _i+1 . be done. At this time, the sensor 23
Which track corresponds to can be determined by which position of the track detection sensor 24 the edge of the banknote 18' passes through.

The banknotes that have been conveyed through the data collection unit 15 are stored in the illustrated storage unit 16 as shown in the illustrated banknote 18″.The data collected during the above-mentioned conveyance is sent from the control unit 17 to the banknote shown in the second figure. The data is transferred to the data editing device 3.

As explained with reference to FIG. 2, in the sample extraction section 101 in the sample data creation section 10,
By introducing the amount of variation, the average value M' and the variance V' are determined for each track and each zone, that is, for each square area 25. In this state, the amount of data becomes enormous, and it is necessary to reduce the number of data so that it is suitable for subsequent processing.

FIG. 5 is an explanatory diagram illustrating the process of obtaining a specimen for simulating the identification rate (acceptance rate).

Now, assume that the data distribution for group A is as shown in FIG. 5A, when the horizontal axis is the distance from the center of distribution for group A. Assuming that the total number of data is N and the number of data to be extracted is S, it is not desirable to disrupt the form of the distribution of the total data. For this reason, as a first aspect,
Divide the range from X ₀ to X ₁ , where data belonging to group A exists, into M equal parts, and each equal distance
The number of data is reduced to S/N for each X _i . That is, since it is expressed as N= _M 〓 ⁱ⁼¹ Y (X _i ), S= _M 〓 ⁱ⁼¹ S/NY (X _i ) is obtained as the data S to be extracted. By doing this, the data belonging to group A can be reduced to S/N.

When focusing only on the existence of group A, the fifth
The mode shown in Figure A is also sufficient. However, group A
If there are groups other than group B, group C, etc., such as group J, even when reducing the number of data mentioned above, the distribution of group A from group B, group C, etc. It is desirable to investigate the state of the data and reduce the number of data so that the distribution state does not fluctuate. For this reason, the second
As an embodiment, the above group B, group C...
Figure 5B shows the distribution of data for group A as seen from
Considering as follows, (i) data Y _AKA of group A corresponding to distance X _K from the distribution center of group A, (ii) data Y _BKA of group A corresponding to distance X _K from the distribution center of group B ,...(x) Data Y _JKA of group A corresponding to distance X _K from the distribution center of group J are totaled to obtain the number of data S to be extracted.

FIG. 6 is an explanatory diagram illustrating the process of obtaining a sample for simulating the misidentification rate.

When simulating misidentification, for example, it is simulated that data belonging to a group other than group A is misclassified as belonging to group A. For this purpose, as a first aspect, FIG.
As shown in the diagram, for other groups, group A
It is sufficient to check the distance from the center of distribution of group A and take samples of data in a range close to group A (for example, the shaded range in the figure). In other words, it is sufficient to delete data without disturbing the distribution of data within the diagonal line. The other groups B and C may also be considered in the same way. The one shown in FIG. 6B corresponds to the second embodiment, in which all groups other than group A are grouped together to obtain a sample.

A discrimination simulation or a misidentification simulation is performed using the sample extracted as described above, but there are variations in the test target discrimination machine, and the sample extracted as described above is used as it is for standard test target discrimination. There is no guarantee that the simulation will be compatible with the machine. Therefore, in order to cancel the influence of the above-mentioned variations, a standard sample is created by introducing a variation amount into the above-mentioned sample. The sensor fluctuation extraction unit 102 shown in FIG. 1 is compatible with this process.

Assuming that a standard sample is obtained as described above, when a simulation is performed using the standard sample, a standard test object discrimination machine will detect 50%, 60%, 70%,
It is necessary to determine the distance from the distribution center of group A to obtain each classification rate such as 80%. The reference value creation unit 103 shown in FIG. 1 corresponds to this process.

When performing the processing in the reference value creation unit 103, for example, the distribution of data belonging to group A is calculated correctly from the entire data in which data belonging to group A, data belonging to group B, etc. are mixed. It is necessary to judge.

FIG. 7 shows an explanatory diagram for explaining the mode of group A determination.

Now, when data belonging to group A and data belonging to group B are mixed, the horizontal axis is the distance from the distribution center of group A, and the vertical axis is the distance from the distribution center of group B. Plot each piece of data. By doing so, the distribution of data belonging to group A and the distribution of data belonging to group B are likely to be divided into two. In the illustrated case, the distribution range of group A is given by condition 1...distance from A, a ₁ ≦a≦ _a, and condition 2... distance from B, b ₁ ≦b≦b ₂ .
If there are other groups C, D, etc., in addition to the above conditions 1 and 2, condition 3...distance from C c ₁ ≦c≦c ₂ condition 4...distance from D The distance d ₁ ≦d≦d ₂ 〓 〓 can be added using AND logic.

Identification rate creation unit/misidentification rate creation unit 1 shown in Figure 1
In step 04, a simulation is performed using the standard specimen obtained as described above. However, when performing a misclassification simulation, taking countermeasures against noise into consideration, from among the misclassification samples obtained as described above, K = (variance of a given group A) / (total variance of group A or group J) In consideration of this, when performing misidentification for group A, perform a simulation by not adding data for one or more square areas where the value of K is small to the distance calculation. You can also do it. By doing so, it is possible to magnify misidentification in the presence of noise, and simulation can be performed in a safer direction.

The logic correction unit 105 shown in FIG. 1 corrects the logic so as to obtain a discrimination logic that can correctly discriminate a data group that causes misclassification from one that does not belong to group A, for example.

As shown in FIG. 8, the paper sheet data evaluation unit 106 shown in FIG. (For example, the voltage value corresponding to the sensed data) changes significantly compared to that of tracks #1 to #10.

FIG. 9 shows a processing flowchart corresponding to the discrimination rate simulation, and the discrimination rate is determined using the value of B/A shown in the figure. Further, FIG. 10 shows a processing flowchart corresponding to a simulation of the misclassification rate, and the misclassification rate is determined using the value of C/A shown in the figure.

〔Effect of the invention〕

As described above, according to the present invention, it is possible to create specimens suitable for various simulations from collected data, and use the results of the simulations to evaluate a test target identification machine.

[Brief explanation of drawings]

Fig. 1 is a diagram showing the principle configuration of the present invention, Fig. 2 is a configuration of an embodiment of the paper sheet validating machine development system of the present invention, Fig. 3 is an example of the configuration of a bill data collector, and Fig. 4 is a diagram of bill data collection. FIGS. 5 to 8 are explanatory diagrams for explaining the operation of the present invention. FIG. 9 is a processing flowchart corresponding to the classification rate simulation. FIG. 10 is a misclassification rate simulation. The corresponding processing flowchart is shown below. In the figure, 1 is a data processing device, 2 is a bill data collector, 3 is a data editing device, 4 is a floppy disk, 6 is a database, 10 is a sample data creation section, 101 is a sample extraction section, and 102 is a sensor The fluctuation extraction unit 103 represents a reference value creation unit.

Claims (1)

[Scope of Claims] 1. A data collector that reads information on set paper sheets including banknotes using a plurality of sensors, and additional data related to the data read by the data collector. a data editing device that adds information, and a data processing device that includes a database that holds information edited by the data editing device and an information processing function unit that processes information in the database. A paper sheet validating machine development system having at least the above information processing function section includes at least a transport state evaluation section that evaluates the validity of the collection state of data collected by the data collector; A database information evaluation unit that checks the data collected by the device, including checking the data acquisition conditions and checking for the inclusion of abnormal data; A specimen data creation unit that creates specimen data to be used for classification classification, and the specimen data created by the specimen data creation unit is used as data that would be collected by the test object classification machine and is used for classification by the test object classification machine. It includes a recognition logic unit that performs simulation, and a discrimination sensor evaluation unit that obtains evaluation information about the sensor of the test object classification machine based on the simulation result by the recognition logic unit, and the sample data creation unit at least , the data collected by the above data collector
Using the data evaluated by the in-base information evaluation unit, a variation amount is introduced in order to correct the deviation in the characteristics of the sensor in the data collector, and then the specimen and/or A sample extraction unit that extracts a sample to be used for simulating a false classification rate; and a sensor that introduces an amount of variation corresponding to a variation range of sensor characteristics in a test object classification machine into the sample extracted by the sample extraction unit. a variation extraction unit, and the sample data creation unit performs the discrimination simulation using a standard specimen resulting from introducing a variation amount by the sensor variation extraction unit. Classification machine development system.