JP2589604B2 - How to distinguish bills and paper sheets - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

The present invention relates to an automatic machine for handling bills such as vending machines, currency exchange machines and automatic transfer / deposit / withdrawal machines, as well as securities, credit securities, tickets, documents, printed matter (hereinafter referred to as "printed matter"). The present invention relates to a banknote used in an automatic machine for handling paper sheets) and a method for discriminating paper sheets.

[0002]

2. Description of the Related Art Conventionally, discrimination between bills and paper sheets has been performed by a bill and discrimination device using various sensors. However, the method of discriminating banknotes and paper sheets using such a sensor has some problems. That is, when discriminating a bill and a sheet based on a signal obtained from the sensor, the state of the device system in which the sensor is assembled, for example, "the bill and the sheet are separated from the sensor as compared with a specified value. Over or too close ", the case where the feed speed of bills and paper sheets is not constant and fluctuates, or the state of bills and paper sheets, for example," wound "and" dirt " Cases such as "wrinkles"
Due to various factors, signals of bills and paper sheets cannot be obtained accurately, and in many cases, bills and paper sheets cannot be distinguished. In order to cope with this situation, methods such as combining a plurality of sensors or patterning the signals obtained from the sensors have been adopted.However, in these methods, in order to correspond to various bills and sheets. Is very strict, which requires a lot of time, and despite the difficulty of adjustment, its recognition rate is limited.

[0003]

SUMMARY OF THE INVENTION In view of the above circumstances, it is an object of the present invention to provide various types of banknotes acquired by sensors and banknotes suitable for discriminating banknotes and sheets based on signals of the sheets. And a method for discriminating paper sheets.

[0004]

A bill acquired by a sensor,
And a primary signal processing unit that processes signals of paper sheets, a neural network threshold value determining unit and a threshold learning unit, a secondary signal processing unit, a neural network bill /
It consists of a paper sheet signal output unit and a banknote / paper sheet learning unit,
In the neural network threshold value learning unit, the secondary signal processing unit learns a threshold value for performing effective and appropriate waveform cutout, and the neural network threshold value determination unit using the threshold value is used for the neural network threshold value determination unit. The waveform data from the primary signal processing unit is input, and the waveform data output by the threshold value determination unit is patterned in the secondary signal processing unit. And learning data of paper sheets, and based on the learning, outputs a paper money signal corresponding to the patterned waveform data from the neural network paper money / paper signal output unit, and a paper sheet signal, and performs determination. The unit determines correctness, type, etc., and displays the determination result on the display unit.

[0005]

[Action] The neural network threshold learning unit learns the various banknotes and paper signals obtained by the sensor,
From this learning, the secondary signal processing unit determines a threshold value for performing effective and appropriate waveform cutout, and based on the threshold value, the neural network threshold value determination unit determines the threshold value of various bills and paper sheets. The signal is processed, and on the other hand, the banknote and the sheet data are learned in the neural network banknote and sheet learning unit. Based on this learning, the secondary signal processing is performed from the neural network banknote and sheet signal output unit. The section outputs signals of banknotes and paper sheets corresponding to the waveform data patterned. This makes it possible to easily and surely process the signals of complicated bills and paper sheets, and easily and reliably discriminate the types of bills and paper sheets.

[0006]

Embodiments of the present invention will be described below with reference to the drawings. FIG. 1 is a configuration diagram of a bill and sheet discriminating method using the present invention, and shows a flow of a series of processes of discriminating bills and sheets. In FIG. 1, reference numeral 1 denotes a sensor unit, which includes various sensors such as a magnetic sensor and an optical sensor or a combination of a plurality of sensors. The sensor unit 1 measures a bill and a sheet signal as analog data. As a signal of a banknote and a sheet, for example, a banknote,
And signals based on the measurement interval between paper sheets and sensors, bills,
And a signal based on the feed speed of the paper sheet, or a signal based on the state of the paper money and the paper sheet, such as “flaw”, “dirt”, and “wrinkle”. An A / D signal converter 2 converts analog data into digital data. Reference numeral 3 denotes a primary signal processing unit for removing noise components of digital data and shaping the waveform. Reference numeral 5 denotes a neural network threshold value determination unit that adjusts an appropriate threshold value for cutting out a waveform in the next process with respect to the strength of the taken-in banknote and sheet signal. As described above, the strength of the signal of the bill and the paper sheet is, as described above, the measurement interval between the bill and the paper sheet and the sensor is too narrow or too wide,
Widely distributed from weak signals to large signals due to the paper bills and paper sheet feeding speed being too fast or too slow, or due to the condition of the bills and paper sheets such as "flaw", "dirt" and "wrinkle". I do. The neural network threshold value determining unit 5 generates a digital waveform for cutting out an effective and appropriate waveform in the next processing for widely distributed bills and paper sheet signals by appropriately adjusting the threshold value. Output data. Reference numeral 6 denotes a neural network threshold learning unit that learns the threshold of the neural network threshold determining unit 5. Details will be described later. 7
Is a secondary signal processing unit which processes the number and strength of digital signals as data from digital waveform data formed by the threshold value of the neural network threshold value determining unit 5. Reference numeral 9 denotes a neural network bill / leaf signal output unit, which takes in the processed data of the secondary signal processing unit 7 and outputs a signal indicating what kind of bill and bill. Numeral 10 is a neural network bill / leaf learning unit, which learns bill and bill data. The processing data of the secondary signal processing unit 7 varies in the number of digital signals and the strength of the digital signals even for the same kind of banknotes and paper sheets, and the determination method of the neural network banknote / paper signal output unit 9 is uniform. In such a case, the recognition rate of bills and paper sheets is low. Details will be described later. 11 is a determination unit,
Based on the output signal of the neural network bill / sheet signal output unit 9, the type of the bill and the kind of the sheet are determined.
Reference numeral 12 denotes a display unit that displays a determination result. In addition, 4
And 8 indicate a learning mode switching unit.

Next, FIG. 2 shows a detailed diagram of the neural network threshold value determining section 5 and the neural network threshold value learning section 6. The neural network threshold determining unit 5 includes an input layer, an intermediate layer, an output layer, and a connection unit. The input layer includes data Us ₁ , Us ₂ , Us ₃ , and Us ₄ output from the primary signal processing unit 3. ,... Usn are input, and this data passes through a connection portion between the input layer and the intermediate layer, a middle layer, a connection portion between the intermediate layer and the output layer, and from the output layer to output data Uo ₁ , Uo ₂ ,.・ ・ Output Uon. Each connection unit is composed of a plurality of connections, and for each connection, learning data (connection coefficient, that is, a threshold value) learned by the neural network threshold learning unit 6 is used. Therefore, the output data of the output layer becomes digital waveform data for performing effective and appropriate waveform cutout in the next processing based on the learned threshold value.
Here, the neural network threshold learning unit 6 will be described. The learning unit 6 includes a neural network having an input layer, an intermediate layer, an output layer, and a connection unit, a comparison unit, and a teacher data supply unit, and learns as follows.
The learning mode switching unit 4 is switched from the determination side to the learning side, and the data Us ₁ , Us ₂ , and Us output from the primary signal processing unit 3
_3, Us _4, enter the · · · Usn to the input layer, the coupling portion of the input layer and the intermediate layer, the intermediate layer, via the connecting portion of the intermediate layer and the output layer, the output data U of the neural network output layer
o _1, Uo _2, and outputs the ··· Uon. On the other hand, the teacher data supply unit supplies teacher signals Ut ₁ , Ut ₂ ,... Utn, compares the output data of the output layer with the teacher signal in the comparison unit, and feeds back the error to each connection. This learning is repeated until the error is within a small allowable range.
In this way, the coupling coefficient (threshold) of each connection is determined. With this learning, an appropriate threshold value for cutting out a waveform in the next processing is obtained for the taken bill signal and the strength of the paper sheet.

Next, the neural network bill / leaf signal output unit 9 and the neural network bill / leaf learning unit 10 will be described. The neural network of the banknote / sheet signal output unit 9 and the banknote / sheet learning unit 10 includes a neural network threshold value determining unit 5 shown in FIG.
Since the configuration and the learning method of the neural network threshold learning unit 6 are the same as those of the neural network threshold learning unit 6, only the differences will be described. Corresponding parts are shown in parentheses. The neural network bill / leaf signal output unit 9 inputs the number and strength data of digital signals output from the secondary signal processing unit 7 to the input layer, and determines the types of bills and bills from the output layer. And output a signal representing the signal. Learning data (coupling coefficients) learned by the neural network banknote / sheet learning unit 10 is used for each connection of the connection units between the layers. Here, the neural network banknote / sheet learning unit 10 inputs the data output from the secondary signal processing unit 7 to the input layer via the learning mode switching unit 8 switched from the output side to the learning side. , The output signal of the neural network is output from the output layer via the intermediate layer. On the other hand, the teacher signal is supplied from the teacher data supply unit, the output signal of the output layer is compared with the teacher signal in the comparison unit, and the error is fed back to each connection between the layers. This learning is repeated until the error is within a small allowable range, and the coupling coefficient of each connection is determined.

Hereinafter, an example in which a 10,000-yen bill is determined according to the present invention will be described. Sensor section (magnetic sensor) 1
Acquires a characteristic part of a 10,000-yen bill as an analog waveform, and converts it into a digital waveform shown in FIG. In FIG. 3, the vertical axis represents the magnetic ink amount of the bill, the horizontal axis represents the relative speed between the magnetic sensor and the bill transport, the upper digital waveform represents the measured value, and the lower one represents the sensor output level. A portion having a large peak value in FIG. 3A shows a characteristic partial waveform of a 10,000-yen bill. In the primary signal processing unit 3, the waveform of FIG. 3A is subjected to noise component removal and waveform shaping, and a waveform as shown in FIG. 3B is output. However, as apparent from FIG. 3B, the bill signal of the 10,000 yen bill includes signals of all intensities from weak signals to large signals. Therefore, the neural network threshold value determining unit 5 uses the threshold value learned by the neural network threshold value learning unit 6 to fetch the waveform of FIG. And outputting digital waveform data for cutting out an appropriate waveform. The secondary signal processing unit 7
As shown in FIG.
It is processed as data on the number and intensity of digital signals that are characteristic of a 100-yen bill. However, even with the same kind of 10,000-yen bill, the number of digital signals and the intensity of the digital signals vary in the processed data of the secondary signal processing unit 7 due to the "dirt" of the bill as described above. Therefore, the neural network banknote / leaf sheet signal output unit 9 uses the coupling coefficient obtained by learning the same kind of banknote data in the neural network banknote / leaf sheet learning unit 10, and even if the input processing data has some variation. That is, if the dirt on the 10,000-yen bill is within the allowable range, it is determined to be a 10,000-yen bill, and a signal representing the 10,000-yen bill is output. This output signal is determined by the determination unit 11 to be a 10,000 yen bill, and the determination result is displayed on the display unit 12.

Although the number of digital signals and the intensity of the digital signals vary in the processed data of the secondary signal processing unit 7 due to the "dirt" of the bill, the measurement interval between the bill and the sensor is too narrow. The same effect can be obtained even if the number of digital signals and the strength of the digital signals fluctuate due to an excessively wide band, a bill feeding speed that is too fast or too slow, or a bill condition such as “flaw” or “wrinkle”. Further, it goes without saying that the present invention can be similarly applied to discrimination of banknotes other than 10,000-yen bills and paper sheets.

[0011]

According to the present invention, the following effects can be obtained. (1) In discriminating bills and paper sheets, a threshold value for effective and appropriate waveform cutout is adjusted using a neural network with respect to sensor signals of complicated bills and paper sheets. Processing of banknotes and paper signals can be performed easily and reliably. (2) It is possible to adapt to discrimination of various bills and paper sheets. (3) Even if the interval between the banknote and the paper sheet and the sensor is not kept strictly constant, and the transport speed of the banknote and the paper sheet fluctuates slightly (may be manual), or Even if the paper sheet has some "flaws", "dirt", "wrinkles", etc., the type of the paper money and the paper sheet can be easily and reliably determined by the function of the neural network.

[Brief description of the drawings]

FIG. 1 is an embodiment of the present invention, and shows a configuration diagram of a method for discriminating bills and paper sheets.

FIG. 2 is a detailed diagram of a neural network threshold determining unit 5 and a neural network threshold learning unit 6 (or a neural network banknote / leaf signal output unit 9 and a neural network banknote / leaf learning unit 10). Is shown.

FIG. 3A shows an output waveform of an A / D signal converter, and FIG.
The output waveform of the secondary signal processing unit and (c) the output waveform of the secondary signal processing unit are shown.

[Explanation of symbols]

 DESCRIPTION OF SYMBOLS 1 Sensor part 2 A / D signal conversion part 3 Primary signal processing part 4, 8 Learning mode switching part 5 Neural network threshold value determination part 6 Neural network threshold value learning part 7 Secondary signal processing part 9 Neural network bills Sheet signal output unit 10 Neural network bill / sheet learning unit 11 Judgment unit 12 Display unit

Claims (1)

(57) [Claims] 1. A sensor for detecting bill and bill signals, a primary signal processing unit for removing noise components and shaping a waveform, a neural network threshold determining unit and a threshold learning unit, and a secondary A signal processing unit, a neural network bill / leaf signal output unit, a bill / leaf learning unit, a determination unit and a display unit, wherein the neural network threshold learning unit includes the secondary signal processing unit. Learning a threshold value for performing effective and appropriate waveform segmentation, inputting the waveform data from the primary signal processing unit to the neural network threshold value determination unit using the threshold value, and The waveform data output by the value determination unit is patterned in the secondary signal processing unit.
And learning the data of paper sheets, based on this learning, the banknote corresponding to the patterned waveform data from the neural network banknotes and paper sheets signal output unit, and output a signal of the paper sheet, the said A method for determining bills and paper sheets, wherein the determination unit determines correctness, type, and the like, and displays the determination result on the display unit.