Currency discrimination device, currency discrimination method, and program

Abstract

Expanding the range of distinguishable currencies. [Solution] The acquisition unit acquires a currency image by capturing a picture of the currency. The first discrimination unit receives an input image of the currency and inputs the currency image into a first trained model that has been trained to output one or more candidate denominations of the currency shown in the image, thereby obtaining candidate denominations of the currency. The second discrimination unit uses a denomination discrimination algorithm for the candidate denomination among the denomination discrimination algorithms for each denomination to determine whether the candidate denomination matches the denomination of the currency.

Description

Technical Field

[0001] The present invention relates to a currency discrimination device, a currency discrimination method, and a program.

Background Art

[0002] Patent Document 1 discloses a technique for optically detecting the surface pattern of a coin to discriminate the denomination of the coin. According to Patent Document 1, the denomination of the coin is tentatively determined from the outer diameter of the coin, and a check is performed according to the tentatively determined denomination to discriminate whether the tentatively determined denomination is correct.

Prior Art Documents

Patent Documents

[0003]

Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0004] However, a currency discrimination device may receive a damaged or deformed currency, and it is not always possible to correctly recognize the outer diameter of the inserted currency. In this case, the inserted currency will be rejected as undiscriminable. An object of the present invention is to provide a currency discrimination device, a currency discrimination method, and a program that can expand the discriminable range of currency.

Means for Solving the Problems

[0005] According to a first aspect of the present invention, the currency discrimination device includes: an acquisition unit that acquires a currency image obtained by photographing a currency; a first discrimination unit that receives an input of an image in which a currency is depicted and inputs the currency image to a first trained model which has been trained to output one or more candidates for the denomination of the currency depicted in the image, thereby obtaining candidates for the denomination of the currency; and a second discrimination unit that uses a denomination discrimination algorithm for the denomination of the currency, among denomination discrimination algorithms for each denomination, to determine whether the candidates for the denomination match the denomination of the currency.

[0006] According to a second aspect of the present invention, the currency discrimination method comprises the steps of: acquiring a currency image obtained by photographing a currency; obtaining candidate denominations of the currency by inputting the currency image into a first trained model that has been trained to accept an input image of a currency and output one or more candidate denominations of the currency shown in the image; and determining whether the candidate denomination matches the denomination of the currency using a denomination discrimination algorithm for the candidate denomination among denomination discrimination algorithms for each denomination.

[0007] According to a third aspect of the present invention, the program causes the computer of a currency processing machine to perform the following steps: acquire a currency image of a currency that has been photographed; input the currency image into a first trained model that has been trained to accept an input image of a currency and output one or more candidates for the denomination of the currency shown in the image, thereby obtaining candidates for the denomination of the currency; and determine whether the candidates for the denomination match the denomination of the currency using a denomination discrimination algorithm for the denomination of the currency, among the denomination discrimination algorithms for each denomination. [Effects of the Invention]

[0008] According to the above embodiment, the currency discrimination device can broaden the range of currencies that can be distinguished. [Brief explanation of the drawing]

[0009] [Figure 1] This is a plan view showing the main parts of the coin processing device according to the first embodiment. [Figure 2]This is a schematic block diagram showing the configuration of the control device according to the first embodiment. [Figure 3] This is a flowchart showing the operation of the coin processing device according to the first embodiment. [Figure 4] This is a flowchart showing the operation of the coin processing device according to the second embodiment. [Figure 5] This is a flowchart showing the security check method by the security check unit according to the third embodiment. [Figure 6] This figure shows an example of the division of a currency image and the determination of the object to be judged according to the first embodiment. [Figure 7] This is a schematic block diagram showing the configuration of a computer according to at least one embodiment. [Modes for carrying out the invention]

[0010] <First Embodiment> Configuration of coin processing device 1 The embodiments will be described in detail below with reference to the drawings. Figure 1 is a plan view showing the main parts of the coin processing device 1 according to the first embodiment. The coin processing device 1 is a coin deposit and withdrawal device that performs a storage process, which counts and sorts loose coins of mixed denominations that are deposited by denomination, and a withdrawal process, which dispenses the stored coins by denomination. Coins are an example of currency. The coin processing device 1 can be used, for example, as a coin change dispenser that constitutes a cash register system. In the following description, "front" refers to the operator's side, "back" refers to the opposite side from the operator, and "left" and "right" refer to the left and right as seen from the operator's side. The coin processing device 1 is an example of a currency discrimination device.

[0011] The coin processing device 1 according to the first embodiment includes a coin dispensing unit 11 and a coin sorting and conveying unit 12. The coin dispensing unit 11 is located on the front side of the coin processing device 1 and separates the inserted coins C one by one and dispenses them toward one side (specifically, toward the left). The coin dispensing unit 11 is located on the front side of the coin processing device 1 and extends in the left-right direction along this front side. The coin dispensing unit 11 has a coin deposit slot 111. The coin deposit slot 111 is located on the opposite side from the coin sorting and transporting unit 12 and accepts loose coins C from outside the coin processing device 1.

[0012] The coin sorting and conveying unit 12 is connected to the coin dispensing side of the coin dispensing unit 11 and receives the coins C dispensed from the coin dispensing unit 11 and conveys them toward the rear. The coin sorting and conveying unit 12 is located on one side (specifically the left side) of the coin processing device 1 and extends in the front-rear direction along this side. Therefore, the coin dispensing unit 11 and the coin sorting and conveying unit 12 intersect with each other, and more specifically, are perpendicular to each other, forming an L-shape overall. The coin sorting and conveying unit 12 has a width that allows the coins C dispensed one by one from the coin dispensing unit 11 to be moved in a single line in the extending direction.

[0013] The coin processing device 1 has multiple coin storage sections 13 located behind the coin dispensing section 11 and below the coin sorting and transporting section 12. The multiple coin storage sections 13 are arranged so as to intersect with the coin sorting and transporting section 12 when viewed from above. Each portion of the coin sorting and transporting section 12 facing each coin storage section 13 is provided with a sorting hole 131 into which coins C can fall, a sorting gate 132 that opens and closes the sorting hole 131, and a solenoid (not shown) that drives the sorting gate 132 to open and close. As a result, when the sorting gate 132 is opened by the solenoid, coins C passing through the coin sorting and transporting section 12 fall through the sorting hole 131 and are stored in the coin storage section 13.

[0014] The coin processing device 1 has a reject passage 14 located behind the coin dispensing section 11 and in front of the coin storage section 13, below the coin sorting and transporting section 12. The reject passage 14 is connected to the coin dispensing opening 15. The portion of the coin sorting and transporting section 12 facing the reject passage 14 is provided with a reject hole 141 through which coins C can fall, a reject gate 142 that opens and closes the reject hole 141, and a solenoid (not shown) that drives the reject gate 142 to open and close. As a result, when the reject gate 142 is opened by the solenoid, coins C passing through the coin sorting and transporting section 12 fall from the reject hole 141 and move through the reject passage 14 to the coin dispensing opening 15.

[0015] 《Control Device》 The solenoid for opening the sorting gate 132 and the solenoid for opening the reject gate 142 are controlled by the control device 30. The control device 30 discriminates the denomination of the coin passing through the coin sorting and conveying unit 12 and opens the sorting gate 132 corresponding to the denomination. When the control device 30 cannot discriminate the denomination of the coin passing through the coin sorting and conveying unit 12, it opens the reject gate 142. An image sensor 51 and a sensor group 52 are provided on the front side of the reject hole 141 in the coin conveying direction in the coin sorting and conveying unit 12. The image sensor 51 is an area sensor or a line sensor and captures an image of the entire coin C. The sensor group 52 is a sensor that detects the characteristics of the coin C used for discriminating the denomination, such as a weight sensor, a thickness sensor, and a material sensor. In FIG. 1, although the image sensor 51 and the sensor group 52 are shown separated, for example, they may be integrally provided at the position of the reference numeral 52 in the drawing.

[0016] FIG. 2 is a schematic block diagram showing the configuration of the control device 30 according to the first embodiment. The control device 30 includes an acquisition unit 31, a candidate determination unit 32, a security check unit 33, a post-processing unit 34, a control unit 35, and a model storage unit 36.

[0017] The acquisition unit 31 acquires a currency image obtained by imaging the coin C from the image sensor 51 and measurement data of the physical characteristics of the coin C from the sensor group 52. The model storage unit 36 stores a coin type discrimination model, which is a machine learning model used for discrimination by the candidate determination unit 32, coin type-specific image feature data used for discrimination by the security check unit 33, and coin type-specific physical feature data used for discrimination by the post-processing unit 34. The coin type discrimination model is trained to take an image in which the coin C appears as input and output the probability distribution of the coin type of the coin C. The learning of the coin type discrimination model will be described later. The image feature data stores the feature amounts of the images of the coin C for each coin type. For example, the image feature data according to the first embodiment is data representing the feature amounts of a plurality of feature points of the coin C determined in advance for each coin type. The physical feature data is data representing the physical features (such as the range of weight, the range of components, the range of thickness, etc.) of the coin C for each coin type.

[0018] The candidate determination unit 32 obtains candidates for the coin type of the currency by inputting the currency image acquired by the acquisition unit 31 into the machine learning model stored in the model storage unit 36. The security check unit 33 compares the currency image acquired by the acquisition unit 31 with the image feature data corresponding to the candidates for the coin type obtained by the candidate determination unit 32, and determines whether the candidates for the coin type match the coin type of the currency. The post-processing unit 34 performs comprehensive identification processing by comparing the measurement data acquired by the acquisition unit 31 with the mechanical feature data corresponding to the coin type determined by the security check unit 33, determines the coin type of the coin C, or determines that the coin type cannot be discriminated.

[0019] The control unit 35 controls the sorting gate 132 or the reject gate 142 based on the discrimination result by the post-processing unit 34. Specifically, when the coin type of the coin C is determined by the post-processing unit 34, the control unit 35 opens the sorting gate 132 corresponding to the coin type when the coin C passes over the coin storage unit 13 corresponding to the coin type. On the other hand, when the coin type of the coin C cannot be discriminated by the post-processing unit 34, the control unit 35 opens the reject gate 142 when the coin C passes over the reject passage 14.

[0020] 《Learning of the Coin Type Discrimination Model》 The coin denomination discrimination model may be composed of, for example, a Convolutional Neural Network (CNN). The input layer of the coin denomination discrimination model is configured to accept images of a predetermined size as input. The output layer of the coin denomination discrimination model has nodes corresponding to each coin denomination. The coin denomination discrimination model is pre-trained to discriminate the denomination of coin C from images in which coin C is pictured. The vector output by the coin denomination discrimination model represents the probability distribution of the denomination of coin C. The learning device (not shown in the diagram) prepares a dataset for training the coin discrimination model, consisting of an image of a coin C (the input sample) and a one-hot vector representing the coin's denomination (the output sample). Images of coin C included in the training dataset may include not only those obtained by photographing actual coin C, but also images obtained by rendering a deformed model—a three-dimensional model of the shape of coin C that has been deformed or partially missing—with the texture of coin C applied to it. Images rendered from a deformed model can be considered images of coins that are missing or deformed. In this way, by including images generated from deformed models in the training dataset, it is possible to identify candidate denominations for coin C that are deformed or missing.

[0021] Operation of coin processing device 1 Figure 3 is a flowchart showing the operation of the coin processing device 1 according to the first embodiment. When a coin C is inserted into the coin dispensing unit 11, the coin dispensing unit 11 dispenses the coin C one by one to the coin sorting and conveying unit 12. The coin sorting and conveying unit 12 conveys the dispensed coins. When a coin C passes below the image sensor 51, the image sensor 51 captures an image (coin image). Once the image sensor 51 captures a coin image, it outputs the coin image to the control device 30. When a coin C passes below the sensor group 52, the sensor group 52 measures physical features. The sensor group 52 outputs the measured data of coin C to the control device 30.

[0022] When the acquisition unit 31 of the control device 30 acquires a coin image from the image sensor 51 (step S1), the candidate determination unit 32 inputs the acquired coin image to the denomination discrimination model (step S2). The candidate determination unit 32 determines the denomination corresponding to the element with the highest probability (confidence) among the probability distribution vectors output by the denomination discrimination model as the candidate denomination for coin C (step S3).

[0023] Next, the security check unit 33 reads image feature data corresponding to the candidate denominations determined in step S3 from the model storage unit 36. The security check unit 33 compares the currency image acquired in step S1 with the image feature data and determines whether the currency image contains a predetermined number or more of the features indicated by the image feature data (step S4). If the number of features indicated by the image feature data contained in the currency image is less than the predetermined number (step S4: NO), the security check unit 33 determines that the denomination is undeterminable (step S5). On the other hand, if the image feature data contains a predetermined number or more of the features indicated by the image feature data (step S4: YES), the security check unit 33 determines that the denomination of coin C is the denomination determined in step S3 (step S6).

[0024] Once the denomination is determined by the security check unit 33 and the acquisition unit 31 acquires measurement data from the sensor group 52 (step S7), the post-processing unit 34 reads the measurement data acquired by the acquisition unit 31 and the mechanical feature data corresponding to the denomination determined in step S6 from the model storage unit 36. The post-processing unit 34 compares the measurement data acquired in step S7 with the mechanical feature data and determines whether the value of the measurement data falls within the range indicated by the mechanical feature data (step S8). If the value of the measurement data falls within the range indicated by the mechanical feature data (step S8: YES), the denomination of coin C is determined. The control unit 35 controls the solenoid to open the sorting gate 132 when coin C passes over the coin storage unit 13 corresponding to the determined denomination (step S9). On the other hand, if some of the measured data values ​​do not fall within the range indicated by the mechanical characteristic data (step S8: NO), the post-processing unit 34 determines that the denomination cannot be determined (step S5).

[0025] If the denomination is determined to be undeterminable in step S5, the control unit 35 controls the solenoid to open the reject gate 142 as the coin C passes over the reject passage 14 (step S10).

[0026] "effect" Thus, according to the coin processing device 1 of the first embodiment, the candidate determination unit 32 uses a trained denomination discrimination model to determine candidate denominations of coin C from a coin image. Since it is difficult to guarantee accuracy in the judgment of a machine learning model, the coin processing device 1 of the first embodiment determines the denomination of coin C by comparing the feature data corresponding to the determined candidate with the coin image. This makes it possible to accurately determine the denomination of coin C even when it is not possible to accurately obtain information such as the outer diameter of coin C. In the coin processing device 1 of the first embodiment, the candidate determination unit 32 is an example of a first discrimination unit that obtains candidate denominations of coins by inputting a coin image into a trained model. The security check unit 33 is an example of a second discrimination unit that uses a denomination discrimination algorithm related to candidate denominations among the denomination discrimination algorithms for each denomination to determine whether the candidate denomination matches the denomination of the coin. In other words, checking a coin image using image feature data corresponding to the denomination is an example of a denomination discrimination algorithm for each denomination. In other embodiments, the control device 30 may not include a security check unit 33, and the post-processing unit 34 may perform denomination discrimination using mechanical feature data corresponding to the candidate denominations determined by the candidate determination unit 32. In this case, the post-processing unit 34 becomes an example of a second discrimination unit that uses a denomination discrimination algorithm for the candidate denomination among the denomination discrimination algorithms for each denomination to determine whether the candidate denomination matches the denomination of the currency. In other words, checking a currency image using mechanical feature data corresponding to the denomination is an example of a denomination discrimination algorithm for each denomination.

[0027] In particular, the denomination discrimination model according to the first embodiment is trained using a training dataset in which images simulating damaged or deformed coins are input samples. As a result, the coin processing device 1 can identify candidate denominations even if the deposited coins C are deformed or damaged, without being unable to distinguish them.

[0028] <Second Embodiment> In the coin processing device 1 according to the first embodiment, the candidate determination unit 32 determines one candidate denomination. Therefore, if the next-choice denomination is the true denomination, the coin processing device 1 according to the first embodiment cannot correctly determine the denomination of coin C and rejects coin C. In contrast, the coin processing device 1 according to the second embodiment is capable of determining multiple candidate denominations. The configuration of the coin processing device 1 according to the second embodiment is the same as that of the first embodiment.

[0029] Operation of coin processing device 1 Figure 4 is a flowchart showing the operation of the coin processing device 1 according to the second embodiment. When the acquisition unit 31 of the control device 30 acquires a currency image from the image sensor 51 (step S21), the candidate determination unit 32 inputs the acquired currency image to the denomination discrimination model (step S22). Based on the probability distribution vector output by the denomination discrimination model, the candidate determination unit 32 identifies multiple denomination candidates and their confidence levels (step S23).

[0030] The security check unit 33 determines the most reliable denomination among the multiple denomination candidates identified by the denomination discrimination model in step S22, and which has not yet been checked by the security check unit 33, as the candidate denomination for coin C (step S24). The security check unit 33 reads the image feature data corresponding to the candidate denomination determined in step S24 from the model storage unit 36. The security check unit 33 compares the coin image acquired in step S21 with the image feature data and determines whether the coin image contains a predetermined number or more of the features indicated by the image feature data (step S25). If the coin image contains a predetermined number or more of the features indicated by the image feature data (step S25: YES), the security check unit 33 determines that the denomination for coin C is the denomination determined in step S24 (step S26).

[0031] On the other hand, if the number of features indicated by the image feature data included in the currency image is less than a predetermined number (step S25: NO), it is determined whether the elapsed time from the start of the denomination discrimination process exceeds a predetermined upper limit time (step S27). The upper limit time may be, for example, the time it takes for the coin C to pass under the image sensor 51 and reach the reject gate 142. If the elapsed time from the start of the denomination discrimination process does not exceed the upper limit time (step S27: NO), the control device 30 returns the process to step S24 and continues the check by the security check unit 33.

[0032] When the denomination is determined by the security check unit 33 and the acquisition unit 31 acquires measurement data from the sensor group 52 (step S28), the post-processing unit 34 reads the measurement data acquired by the acquisition unit 31 and the mechanical feature data corresponding to the denomination determined in step S26 from the model storage unit 36. The post-processing unit 34 compares the measurement data acquired in step S28 with the mechanical feature data and determines whether the value of the measurement data falls within the range indicated by the mechanical feature data (step S29). If the value of the measurement data falls within the range indicated by the mechanical feature data (step S29: YES), the denomination of coin C is determined. The control unit 35 controls the solenoid to open the sorting gate 132 when coin C passes over the coin storage unit 13 corresponding to the determined denomination (step S30). On the other hand, if some of the measured data values ​​do not fall within the range indicated by the mechanical characteristic data (step S29: NO), or if the elapsed time from the start of the denomination determination process in step S26 exceeds the upper limit time (step S26: YES), the post-processing unit 34 determines that the denomination cannot be determined (step S31).

[0033] If the denomination is determined to be undeterminable in step S31, the control unit 35 controls the solenoid to open the reject gate 142 as the coin C passes over the reject passage 14 (step S32).

[0034] "effect" Thus, according to the coin processing device 1 of the second embodiment, the candidate determination unit 32 distinguishes between multiple denominations of coins based on image feature data, in order of increasing confidence. As a result, the coin processing device 1 can distinguish between multiple denominations of coins before the coin C reaches the reject gate 142, thereby expanding the range of distinguishable coins.

[0035] <Third Embodiment> In the third embodiment, the control unit 35 identifies the rotation angle of the coin C based on the coin image, corrects the coin image based on the rotation angle, and performs a security check.

[0036] Position determination model The model storage unit 36 ​​according to the third embodiment stores a position determination model for each coin denomination. The position determination model is a trained model that receives an input image showing a part of a coin C and outputs the position of the part of the coin C that is shown in the image. The position of the coin C may be expressed, for example, as coordinates in an image in which the coin C is shown in its upright position. The position determination model may be composed of, for example, a CNN. The input layer of the position determination model is configured to accept images of a predetermined size obtained by dividing the coin image into a predetermined number of parts. The output layer of the position determination model has nodes corresponding to each position of coin C. The denomination discrimination model is pre-trained to determine the position of a part of coin C from an image in which only a part of the coin C is visible. The vector output by the denomination discrimination model represents the probability distribution of the position of coin C. The learning device (not shown in the diagram) prepares a training dataset for the position determination model, which consists of an image showing a portion of a coin C (the input sample) and a one-hot vector representing the position (the output sample).

[0037] Security Check Method In the third embodiment, the security check unit 33 determines the rotation angle of the coin C in the following procedure in step S4 of the first embodiment or step S25 of the second embodiment. Figure 5 is a flowchart showing the security check method by the security check unit 33 of the third embodiment. Figure 6 is a diagram showing an example of the division of the currency image and the determination of the object to be judged according to the first embodiment.

[0038] First, the security check unit 33 divides the currency image into multiple sub-images (step S41). The security check unit 33 then extracts from the sub-images those with a blank area (area with low contrast) that is smaller than a predetermined percentage as targets for determination (step S42). This is because sub-images with a large blank area lack features and are therefore highly likely to be misidentified by the position determination model.

[0039] The security check unit 33 inputs the extracted partial images of the items to be judged into the position determination model to obtain a vector representing the probability distribution of the position of the coin C corresponding to each partial image (step S43). This allows the security check unit 33 to identify the position where the features appearing in the partial image appear on the coin C in the correct position. For each partial image of the items to be judged, the security check unit 33 calculates the rotation angle from the relationship between the position of the features appearing in the partial image on the currency image and the position on the coin C in the correct position (step S44). The rotation angle may be calculated, for example, according to a table that calculates the rotation angle from two positions in advance, or it may be calculated by performing a rotation calculation on the two positions. The security check unit 33 identifies the median of the rotation angles obtained from each partial image as the rotation angle of the coin C (step S45).

[0040] The security check unit 33 rotates and corrects the coin image at a specified rotation angle and compares it with image feature data to determine the denomination (step S46).

[0041] "effect" According to the third embodiment, the security check unit 33 identifies the tilt of the coin C in the currency image and determines whether a candidate denomination matches the denomination of the currency based on the tilt of the coin C. The security check unit 33 can improve the accuracy of denomination determination by correcting the tilt of the coin C by rotation.

[0042] In the third embodiment, the security check unit 33 determines the rotation angle using the position with the highest confidence level from the vector representing the probability distribution, but is not limited to this. For example, in other embodiments, the security check unit 33 may determine the rotation angle for all positions corresponding to each element of the vector and determine a single rotation angle by a weighted average based on confidence levels.

[0043] In the third embodiment, the rotation angle is determined for each of the partial images to be judged in step S44, but this is not limited to this, and calculations may be performed for only some of the partial images.

[0044] In other embodiments, without performing rotation correction, a security check may be performed using the partial image estimated in step S43 to contain the feature points.

[0045] <Other Embodiments> Although one embodiment has been described in detail above with reference to the drawings, the specific configuration is not limited to that described above, and various design changes are possible. In other embodiments, the order of the above-described processes may be changed as appropriate. Also, some processes may be executed in parallel. The control device 30 of the coin processing device 1 according to the above embodiment may be composed of a single computer, or the configuration of the control device 30 may be divided among multiple computers, and the multiple computers may cooperate with each other to function as the control device 30. In this case, some of the computers that make up the control device 30 may be mounted on other cash processing machines.

[0046] The currency discrimination device according to the above embodiment is implemented in a coin processing device 1 that sorts coins C, but is not limited to this. For example, currency discrimination devices according to other embodiments may be implemented in a bundling machine or a cash deposit / dispensing machine. When the currency discrimination device is installed in a bundling machine, only coins of a specified denomination are sent to the accumulation unit, and after accumulating a predetermined number of coins, the bundling and packaging process is performed, and coins of denominations other than the specified denomination are sorted to a part different from the rejection unit. In other embodiments, the currency discrimination device may distinguish the denomination of banknotes instead of coins C.

[0047] The coin processing device 1 according to the above embodiment guides all coins C whose denomination has been successfully identified to the coin storage unit 13, but is not limited to this. For example, the coin processing device 1 according to another embodiment attempts to identify the denomination using a conventional method (for example, a method of determining candidate denominations based on the size of the outer diameter), and guides those that have been successfully identified to the coin storage unit 13 corresponding to that denomination. On the other hand, coins that have failed to be identified by the conventional method but have been successfully identified by the method according to the above embodiment may be guided to a soiled coin storage unit corresponding to that denomination, which is provided separately from the coin storage unit 13.

[0048] <Computer Configuration> Figure 7 is a schematic block diagram showing the configuration of a computer according to at least one embodiment. The computer 90 includes a processor 91, main memory 92, storage 93, and an interface 94. The control device 30 of the coin processing device 1 described above is implemented in the computer 90. The operation of each processing unit described above is stored in storage 93 in the form of a program. The processor 91 reads the program from storage 93, loads it into main memory 92, and executes the above processing according to the program. The processor 91 also allocates memory areas in main memory 92 corresponding to each of the above-mentioned memory units according to the program. Examples of the processor 91 include a CPU (Central Processing Unit), a GPU (Graphics Processing Unit), and a microprocessor.

[0049] The program may be for implementing some of the functions that the computer 90 is to perform. For example, the program may perform functions in combination with other programs already stored in storage, or in combination with other programs implemented on other devices. In other embodiments, the computer 90 may include a custom LSI (Large Scale Integrated Circuit) such as a PLD (Programmable Logic Device) in addition to, or instead of, the above configuration. Examples of PLDs include PAL (Programmable Array Logic), GAL (Generic Array Logic), CPLD (Complex Programmable Logic Device), and FPGA (Field Programmable Gate Array). In this case, some or all of the functions implemented by the processor 91 may be implemented by the integrated circuit. Such an integrated circuit is also included as an example of a processor. In other embodiments, the computer 90 may be virtualized on one or more computers.

[0050] Examples of storage 93 include magnetic disks, magneto-optical disks, optical disks, and semiconductor memory. Storage 93 may be an internal medium directly connected to the bus of the computer 90, or it may be an external medium connected to the computer 90 via an interface 94 or a communication line. Furthermore, if this program is delivered to the computer 90 via a communication line, the computer 90 that receives the delivery may load the program into the main memory 92 and execute the above processing. In at least one embodiment, storage 93 is a tangible storage medium that is not temporary.

[0051] Furthermore, the program may be intended to implement some of the functions described above. In addition, the program may be a so-called differential file (differential program) that implements the functions described above in combination with other programs already stored in storage 93. [Explanation of Symbols]

[0052] 1...Coin processing unit 11...Coin dispensing unit 111...Coin deposit slot 12...Coin sorting and transport unit 13...Coin storage unit 131...Sorting hole 132...Sorting gate 14...Reject passage 141...Reject hole 142...Reject gate 15...Dispensing slot 30...Control device 31...Acquisition unit 32...Candidate determination unit 33...Security check unit 34...Post-processing unit 35...Control unit 36...Model memory unit 51...Image sensor 52...Sensor group 90...Computer 91...Processor 92...Main memory 93...Storage 94...Interface C...Coin

Claims

1. An acquisition unit that acquires a currency image by photographing the currency, A first discriminant unit obtains candidate denominations of currency by inputting the currency image to a first trained model that has been trained to accept an image of currency as input and output one or more candidate denominations of currency shown in the image. A second discrimination unit that uses a denomination discrimination algorithm for the candidate denomination among the denomination discrimination algorithms for each denomination to determine whether the candidate denomination matches the denomination of the currency, A currency identification device equipped with the following features.

2. The first pre-trained model was trained using images simulating missing or deformed coins. The currency discrimination device according to claim 1.

3. The first discrimination unit outputs a plurality of candidate denominations and the confidence level of each candidate. The second discrimination unit determines, in order of increasing reliability, whether the candidate denomination matches the denomination of the currency. The currency discrimination device according to claim 1.

4. The aforementioned denomination discrimination algorithm is: To identify the tilt of the coins shown in the aforementioned coin image, Based on the tilt of the coin, determine whether the candidate denomination matches the denomination of the coin. The currency discrimination device according to claim 1, including the following:

5. The aforementioned denomination discrimination algorithm is: A second trained model, which is trained to accept an image showing a portion of a coin of the aforementioned denomination and output the position of the portion shown in the image, is input with multiple partial images obtained by dividing the coin image, thereby obtaining the position of the coin shown in the multiple partial images. From the position of the coin corresponding to the plurality of partial images, the tilt of the coin as depicted in the coin image is determined. The currency discrimination device according to claim 4, including the following:

6. The steps include obtaining a currency image by photographing the currency, The steps include: obtaining candidate denominations of the currency by inputting the currency image into a first trained model that has been trained to accept an image of currency as input and output one or more candidate denominations of the currency shown in the image; Among the denomination discrimination algorithms for each denomination, the step of determining whether the candidate denomination matches the denomination of the currency, using the denomination discrimination algorithm for the candidate denomination, A currency identification method comprising the following features.

7. In the currency processing machine's computer, The steps include obtaining a currency image by photographing the currency, The steps include: obtaining candidate denominations of the currency by inputting the currency image into a first trained model that has been trained to accept an image of currency as input and output one or more candidate denominations of the currency shown in the image; Among the denomination discrimination algorithms for each denomination, the step of determining whether the candidate denomination matches the denomination of the currency, using the denomination discrimination algorithm for the candidate denomination, A program to execute.

Images