JPH04118796A - Paper money discriminator - Google Patents

Abstract

PURPOSE:To easily obtain a paper money discrimination system with high accuracy by converting the distinctive print pattern of a paper money to an electrical signal waveform, performing conversion processing to the input pattern of a neural net, and performing discrimination by inputting the pattern to the neural net. CONSTITUTION:A magnetic head 4 scans the prescribed area of the paper money 1 according to the travel of the paper money 1, and since the paper money 1 is printed by using magnetic ink, the light-and-shade level of print on a scanning area can be taken out as the electrical signal waveform. The output of the magnetic head 4 is amplified by an amplifier circuit 5, and furthermore, it is converted to a digital value by an A/D conversion circuit 6, and input to the neural net 8 can be performed in parallel, therefore, the value is stored once in memory 7, and after that, it is inputted to an input processing means 8, and is conversion-processed to the input pattern of the neural net, then, is inputted to the neural net. Thereby, a decision signal representing to which paper money out of plural kinds of patterns learned it corresponds or whether it is a true or a false paper money is outputted, which obtains superior decision accuracy.

Description

[Detailed description of the invention] <Industrial application field> The present invention relates to a banknote validating device.

<Prior Art> As an example of a banknote discrimination method in a conventional banknote validation device, the characteristic points of various banknotes, such as the density level of printing and the dimensions of banknotes, are determined based on regular banknotes established as standards for each banknote. Along with setting the range,
A determination range regarding the feature points is set from the assumed counterfeit banknote, and both determination is made as to whether or not the banknote to be discriminated is within the determination range of a genuine banknote, and whether or not it is not within the determination range of a counterfeit banknote. The type and authenticity of banknotes can be determined by performing this process.

<Problem to be solved by the invention> However, in the conventional method, as described above, it is necessary to set a judgment range for distinguishing the type of banknote and a judgment range for distinguishing the authenticity of the banknote. , and it was difficult to accurately set the re-judgment range.

In other words, even if it is a genuine banknote, there may be variations in printing or dimensions, and old banknotes may have wrinkles or dirt, so this needs to be taken into account.Also, on the validation device side, the banknote transport system, amplifier, etc. Is it necessary to consider variations in circuits, environmental characteristics, aging, etc.?

For this reason, if the judgment range is set too strictly without considering the above-mentioned variation factors, even genuine banknotes will be judged as counterfeit banknotes more frequently, while if the judgment range is set roughly in consideration of the above-mentioned factors, the opposite will occur. The frequency with which counterfeit banknotes are judged as genuine banknotes increases. Therefore, it is difficult to accurately set the determination range, and there is a limit to the discrimination accuracy.

The present invention has been made in view of the above circumstances, and an object of the present invention is to provide a banknote validating device that does not require difficult determination ranges for characteristic points of banknotes and has excellent determination accuracy.

<Means for Solving the Problems> Therefore, the present invention provides a conversion means for converting the printing pattern of a banknote to be discriminated into an electrical signal, and a conversion means for converting the electrical signal outputted from the conversion means into an input pattern of a neural network. The input processing means to process and the characteristics of the printing patterns of plural types of banknotes to be discriminated are learned in advance, and the input processing means inputs the output of the input processing means, and the input printing patterns are learned in advance. The present invention includes a neural network that outputs a determination signal indicating which of the banknotes to be discriminated corresponds to the type and the authenticity of the banknote.

<Operation> In such a configuration, whether the printing pattern of the banknote to be identified or
It is converted into an electrical signal by a conversion means.

Next, the electrical signal corresponding to the print pattern outputted from the converting means is input to the input processing means to be converted into an input pattern for the neural network, and then input to the neural network. The neural network is trained in advance about multiple types of banknotes to be discriminated, and by processing the input signal according to the learning, it is possible to determine whether the printing pattern of the banknote to be discriminated or one of the multiple types of banknotes that has been learned. A determination signal indicating which banknote falls under this category and whether the banknote is a genuine banknote or a counterfeit banknote is output.

That is, by having the neural network learn the printing patterns of each banknote to be discriminated in advance so that it can determine the type and authenticity of the banknote when inputting a processing signal corresponding to the print pattern of the banknote to be discriminated, The type and authenticity of banknotes can be determined simply by inputting print pattern signals to a neural network without setting difficult determination ranges for each banknote to be distinguished.

<Example> Hereinafter, one embodiment of the present invention will be described based on the drawings.

In FIG. 1, a banknote conveying section 3 comprising a pair of upper and lower rollers 2, etc., which nip and convey the inserted banknote 1, converts the printed pattern of the banknote 1 into an electric signal at a predetermined position within the banknote conveyance path. A magnetic head 4 is arranged as a converting means. Thereby, the magnetic head 4 scans a predetermined area of the banknote 1 as the banknote 1 moves. Since the banknote 1 is printed using magnetic ink, the density level of the printing on the scanning area can be extracted by the magnetic head 4 as an electric signal waveform (see A in FIG. 2).

Here, the magnetic head 4 is installed at a position where it can scan areas where the printing patterns of the various banknotes to be discriminated are largely different. Further, it is preferable that a plurality of magnetic heads 4 are arranged in parallel in a direction perpendicular to the conveyance path of the banknote 1 to scan a plurality of areas. Note that an optical image sensor may be used instead of the magnetic head 4 to detect the gray level of the banknote l.

The output of the magnetic head 4 is amplified by an amplifier circuit 5, further converted into a digital value by an A/D conversion circuit 6, and then stored in a memory 7 in order to input it in parallel to a neural network 8, which will be described later. do. In this case, as shown in FIGS. 3 and 4, the amplified output is sampled at predetermined sampling intervals t1, and the sampling value V
, is A/D converted, and its digital value S is sequentially stored in each address a1 of the memory 7. Then, the digital values stored in each address ai of the memory 7 are made to correspond to each unit of the input layer of the neural network 8. Therefore, the A/D conversion circuit 6 and memory 7 constitute input processing means.

After A/D conversion, the gray level signal of the banknote 1 stored in the memory 7 is input to the neural network 8, and the calculation by the neural network 8 determines which of the plurality of banknotes to be discriminated corresponds to the corpse. Otherwise, a determination signal indicating the authenticity of the banknote is output, and it is determined whether the banknote is 1 or not.

Here, I will briefly explain neural networks.

A neural network is a network that imitates the human brain, and is made up of multiple units that correspond to neurons in the brain and are connected in a complex way. Deciding (learning)
It is possible to embed pattern recognition functions and knowledge processing functions, for example, "Nikkei Electronics J 1987, 8
It is introduced in pages 115 to 124 of the 10th issue of the month (Nα427).

First, the structure of the unit modeled as a neuron is shown in the fifth section.
As shown in the figure. Unit (Uron model) Ui converts the sum of inputs Oj from other units according to a certain rule and outputs O1, but each connection with other units has a variable weight Wij (synaptic weight). ing.

This weight Wij is used to express the strength of the connection between each unit, and changing this value will substantially change the structure of the neural network without changing the connection state.

Learning of the neural network involves changing this value, and the weight WiJ takes positive, zero, or negative values.

When inputs are received from a certain unit Ui or a plurality of other units, and the sum of the inputs is expressed as NET, the sum of the inputs of the unit Ui is NETi=ΣWij·Oj.

Each unit Ui applies this input summation NET to the function f and converts it into an output O1 as shown in the following equation.

0i=f (NETi) =f(ΣWij-Oj) This interval f may be different for each unit Ui, but generally the sigmoid function shown in FIG. 6 is used.

This sigmoid function is a differentiable pseudo-linear function, and can be expressed as fi = 1 + e """'.The range is 0 to 1, and as the input value increases, it approaches 1, and as it decreases, it approaches O. When , it becomes 0.5.In some cases, by adding a threshold value θ (bias), it becomes 0.5.

FIG. 7 is a diagram showing an example of a network structure, and shows a hierarchical structure consisting of an input layer, an intermediate layer, and an output layer. Each unit is connected from the input layer to the output layer, but units within each layer are not connected to each other. In addition, the in-kani knit and the out-kani knit are independent.

In such a neural network, when input data is given to each unit in the input layer, this signal is transformed in each unit, transmitted to the intermediate layer, and finally comes out from the output layer, or in order to obtain the desired output. It is necessary to set the strength of the connection between each unit, that is, the weight, to an appropriate value. This weight setting is performed by training the neural network as follows.

First, all weights are set randomly, and input data for learning (data whose desired output is known in advance) is given to each unit in the input layer. At this time, the output value from each unit of the output layer is compared with the desired output value, and the value of each weight is sequentially corrected from the output layer side so as to reduce the difference (error). This is then repeated using a large amount of learning data until the error converges.

As shown in FIG. 2, the neural network 8 of this embodiment is composed of an input layer, an intermediate layer, and an output layer, and the input layer has units U3. U 2...U! It consists of
Each unit U1. U2. . . . A signal value of a corresponding gray level is inputted to U8.

Each unit U + , U 2 , . . . , Ua of the intermediate layer is coupled to all of the units U + , U t , . Note that the number of units in the intermediate layer is determined empirically. In addition, as in this example, in addition to the one-layer structure, two-layer structure is also possible.
A multilayer structure such as a layer or three layers may be used.

Each unit U,, U2 . . . U2 of the output layer is connected to each unit U, , U2 . ...All are combined with Ua. The number of units in the output layer is equal to the number of banknotes to be discriminated plus one unit for determining authenticity of banknotes.

Then, in order to distinguish the printing pattern of the inserted banknote 1 or the banknote when inputted to the neural network 8, the printing pattern of various banknotes to be judged is learned in advance. That is, the printing patterns known in advance for various banknotes are input to the neural network 8, and the input data is input to the neural network 8 while changing the input data so that the type of banknote and the judgment signal of the neural network 8 match. The weights between units are corrected sequentially from the output layer side to gradually increase the accuracy rate of judgments. Through such learning, when the printing pattern of a certain banknote is input, only the output of the unit of the output layer corresponding to that banknote becomes 1.

As shown in FIG. 2, the signal obtained by converting the analog electric signal waveform of the gray level of the inserted banknote l is input to the neural network 8 trained in advance in this way. It is possible to distinguish the types of banknotes and the authenticity of banknotes with high accuracy. Further, there is no need to set a difficult determination range for banknote discrimination. Furthermore, even if the classification target changes, it can be easily handled by learning accordingly.

That is, in banknote discrimination using the neural network 8, if each printing pattern of the banknote to be judged is determined in advance and the discrimination ability required for such printing pattern is set,
All that is left to do is decide on the unit configuration of the neural net 8, and transfer the sample printing pattern data to the neural net 8.
By inputting information into the computer and learning the system so that only the desired answer comes out, banknotes can be discriminated with high accuracy.

<Effects of the Invention> As explained above, according to the present invention, the characteristic printing pattern of banknotes is converted into an electrical signal waveform, which is converted into an input pattern of a neural network and inputted into the neural network. Since we have made it possible to differentiate banknotes, there is no need to set difficult judgment ranges to identify banknotes.
A highly accurate banknote discrimination system can be easily constructed. Furthermore, even if the classification target changes, it can be easily handled by simply performing additional learning.

[Brief explanation of drawings]

Fig. 1 is a system configuration diagram showing one embodiment of the present invention;
The figure is a schematic diagram showing the configuration of the neural network of the same embodiment as above, Figure 3 is a diagram explaining the digital conversion operation of an analog electric signal waveform, Figure 4 is a diagram showing the memory storage state of digital values, and Figures 5- Figure 7 is a diagram for explaining a neural network. Figure 5 is a block diagram of a unit, Figure 6 is a diagram showing an example of the input/output characteristics of a unit, and Figure 7 is a diagram showing the structure of a general neural network. FIG. 1...Banknote 3...Banknote transport section 4...Magnetic head 5...Amplification circuit 6...A/D conversion circuit 7...Memory 8 Neural net

Claims (1)

[Claims] A converting means for converting a printing pattern of a banknote to be discriminated into an electrical signal, an input processing means for converting the electrical signal output from the converting means into an input pattern of a neural network, and printing of multiple types of banknotes to be discriminated. The characteristics of the pattern are learned in advance, and the output of the input processing means is input, and it is determined which of the plurality of types of banknotes to be discriminated that the input printing pattern corresponds to and the authenticity of the banknote. What is claimed is: 1. A banknote validating device comprising: a neural network that outputs a determination signal indicating that the banknote is valid;