Computer-implemented method, device and system for verifying at least one character string applied to at least one object

Abstract

The invention relates to a computer-implemented method (100) for verifying at least one character string (14) applied to at least one object (10), in particular at least one container, which has characters at different positions, in particular a coding, wherein the character string (14) is arranged on at least one section (12) of the object (10) and has been produced according to at least one template character string (22, 24), comprising at least the following steps: receiving (102) an input signal indicating image information (16) of at least the section (12); determining (104) at least one evaluation value for the character string (14) displayed in the image information (16) by means of at least one trained adaptive algorithm (26), wherein the evaluation value indicates at least one probability of an occurrence of the template character string (22, 24) in the image information (16); and providing (106) at least one output signal, indicating the at least one evaluation value, for verifying the character string (14) on the object (10).

Description

[0001] Computer-implemented method, device and system for verifying at least one string of characters affixed to at least one object

[0002] The invention relates to a computer-implemented method, a device and a system for verifying at least one string of characters affixed to at least one object, in particular at least one container.

[0003] Objects, especially food containers, can be marked with imprints, embossing, and / or labels to display information such as batch numbers and / or best-before dates. This information is then inspected for verification. The information on the objects can be captured using inspection equipment and subjected to character recognition (CR). Character recognition can, for example, be used to identify letters or numbers.

[0004] For the recognition of the characters, it is known, for example from CN 13989806 A, to use a Connectionist-Temporal-Classification model.

[0005] The object of the invention is to provide a computer-implemented method for verifying at least one string of characters applied to at least one object, which exhibits increased robustness in the verification of the string.

[0006] The problem is solved by the features of the independent claims. Advantageous further developments are the subject of the dependent claims and the following description.

[0007] The invention relates to a computer-implemented method for verifying at least one string of characters affixed to at least one object, in particular at least one container, which has characters at different positions, in particular an encoding, wherein the string of characters is arranged on at least one section of the object and was produced according to at least one template string of characters, comprising at least the following steps: receiving an input signal that displays image information of at least the section; determining at least one evaluation value for the string of characters displayed in the image information by means of at least one trained machine learning algorithm, wherein the evaluation value indicates at least one probability for the occurrence of the template string in the image information;and providing at least one output signal indicating the at least one evaluation value to verify the string on the object.

[0008] The computer-implemented method thus determines the probability of a given template string being present, instead of explicitly reading each character of the string. This eliminates the need for an explicit reading, such as a subsequent comparison of the read characters with the characters of the template string. Instead, the probability of the characters being present in the template string is determined directly from the values ​​obtained by the trained, machine learning algorithm. The template string may, for example, have been used to apply the string to the object, such as by printing, embossing, and / or labeling.The trained machine learning algorithm can take the image information of the section as input data and output at least one score indicating the probability of the template string occurring in the image information. The trained machine learning algorithm can be further trained to output a probability for each character in a given character set that can occur in a string, indicating whether the character is present in the image information. The character set can include, for example, characters from an alphabet, numbers, and / or punctuation marks, especially spaces, or other characters. By using a machine learning algorithm trained with sufficiently diverse data, no configuration or training of specific characters or fonts is necessary. Furthermore, no segmentation or...Preprocessing of the image data is performed, allowing for simple parameterization. Furthermore, the trained machine learning algorithm can be provided with character tuples, for example, of similar characters, so that these can be assessed for similarity. For example, the character tuple could contain "S" and "5", or "0" and "0", or "1", "I", and "I". Verification of at least one character string can thus be performed with increased robustness, since the probability of the presence of the template character string determined in this way is independent of possible character confusions.

[0009] The probability explained above is not necessarily a statistical probability. It could be a probability estimated by the machine learning algorithm.

[0010] According to some embodiments, it is conceivable that the trained learning algorithm can comprise at least one artificial neural network, in particular a Convolutional Neural Network, into which the image information is input and which outputs the at least one evaluation value.

[0011] The artificial neural network can be designed as a Convolutional Neural Network, which has been trained to recognize the positions of characters in a string from the image information and to specify a probability value for each position as to whether a particular character, for example a character of an alphabet or a number or punctuation mark, is present at that position.

[0012] Furthermore, the artificial neural network can be a Connectionist-Temporal Classification (CTC) model. This can be done particularly efficiently with the CTC model, as precisely aligned example images are not required for training the model.

[0013] According to some embodiments, it is conceivable that the trained learning algorithm can output at least a matrix of probabilities for finding a predefined character at a specific position in the string.

[0014] For example, by multiplying or otherwise combining the probabilities at the different positions for specific characters, an evaluation value can be determined that provides a probability for a particular string of characters. This allows an evaluation value to be determined for a large number of different strings using simple means. The string with the highest evaluation value can be output in further examples.

[0015] According to some embodiments, it is conceivable that before determining the assessment value, at least one region of interest can be defined in the section displayed in the image information, wherein the region contains the string.

[0016] According to some embodiments, it is conceivable that in the step: Determining at least one assessment value, only the image information with which the region can be displayed can be used.

[0017] In this way, the amount of image information can be limited to the region, thus reducing the amount of information and speeding up the process.

[0018] According to some examples, it is conceivable that if several regions are defined, a separate assessment value can be determined for each region.

[0019] For example, if the string has multiple rows or columns, providing a region allows you to select a row or column that can be examined independently of the others. Further regions can then be provided for the remaining rows or columns, enabling subsequent examination. This improves the probability calculation.

[0020] According to some embodiments, the template string could have a predefined template string format, and the trained machine learning algorithm could determine the probability of a string occurring within that format. For example, if the string represents a batch number, the template string could have five positions, four of which could contain two pairs of numbers separated by a period or a slash. To verify the batch number, it might be sufficient to detect that two pairs of numbers are represented, separated by a period or a slash. In this case, there would be a high probability of the template string with the template string format occurring in the image information.Instead of reading each individual character of the string and comparing it to the characters of the intended string, and issuing an error message or "not OK" message in case of differences, the string that most likely matches the template string format and is contained within the section can be output, and a probability can be calculated for it. Depending on whether the probability of a correctly formatted string occurring is still high enough, the string verification can be performed accordingly. This further improves the robustness of the verification process.

[0021] The template string format can define whether a number format, letter format, etc., should be used. The template string format can apply to every position or only to a subset of positions in the string.

[0022] According to some embodiments, it is conceivable that before the step of determining an assessment value, at least one orientation of the image information with the character string can be determined and at least partially adapted to at least one orientation of the template character string.

[0023] For example, if the template string only contains straight lines of characters and the string was applied to a curved surface and therefore appears curved in the image information, aligning the image information can prevent the trained machine learning algorithm from underestimating the probability of the template string being present in the image information due to misalignment.

[0024] According to some embodiments, it is conceivable that a most probable string can be determined based on the at least one evaluation value and provided as a further output signal.

[0025] For example, if a best-before date is represented by the string, the best-before date can be output in addition to the assessment value. This allows for double verification of the string.

[0026] The invention further relates to a computer program product comprising instructions which, when the program is executed by a computer, cause it to perform the steps of the method according to the preceding description.

[0027] The advantages, effects, and further developments of the computer program product result from the advantages, effects, and further developments of the method described above. Therefore, reference is made to the preceding description in this regard. A computer program product can be understood, for example, as a data carrier on which a computer program element is stored, containing instructions executable by a computer. Alternatively or additionally, a computer program product can also be understood, for example, as a permanent or volatile data storage medium, such as flash memory or main memory, that contains the computer program element. However, this does not exclude other types of data storage media that contain the computer program element.

[0028] The invention further relates to a device for verifying at least one string of characters attached to at least one object, in particular at least one container, wherein the string of characters is arranged on at least one section of the object and was produced according to at least one template string, comprising at least one processor unit, an inspection device and an inspection area for at least one object, with at least one string of characters to be inspected, wherein the inspection device is configured to determine image information of at least one section of an object arranged in the inspection area and to provide the image information, wherein the processor unit is signal-connected to the inspection device by means of a signal connection, wherein according to the invention it is provided that the processor unit is configured to carry out the method according to the preceding description.

[0029] The advantages, effects, and further developments of the device result from the advantages, effects, and further developments of the method described above. To avoid repetition, reference is therefore made to the preceding description in this regard.

[0030] The invention further relates to a system for manufacturing and / or treating objects, comprising at least one device for transporting at least one object to be treated along a transport direction, a device for applying at least one string of characters to at least one section of an object, which is arranged on the device for transporting, and at least one device according to the preceding description, which is arranged in the transport direction behind the device for applying a string of characters to the device for transporting, wherein the device for transporting extends through the inspection area.

[0031] The advantages, effects, and further developments of the system result from the advantages, effects, and further developments of the device and method described above. To avoid repetition, reference is therefore made to the preceding description in this regard.

[0032] The invention is described below with reference to an exemplary embodiment and the accompanying drawing. Figure 1 shows a flowchart of the process;

[0033] Figure 2a, b shows a schematic representation of a container with a string of characters;

[0034] Figure 3a, b shows a schematic representation of a template string;

[0035] Figure 4 shows a schematic representation of a trained, machine learning algorithm;

[0036] Figure 5 shows a schematic representation of the region read by the algorithm;

[0037] Figure 6 shows a schematic representation of a matrix; and

[0038] Figure 7 is a schematic representation of a plant.

[0039] In Figure 1, a computer-implemented method for verifying at least one string of characters affixed to at least one object is designated in its entirety by the reference numeral 100.

[0040] Figure 2a shows an object 10, which could be, for example, a can. The string 14 may be affixed to a section 12 of the object 10, which could, for example, be a concave can bottom. The string 14 could, for example, have been printed on the section 12.

[0041] The string 14 was produced according to a template string 22, 24. The template string 22, 24 can, for example, specify the general format of the string 14. If the string 14 represents, for instance, a sequence of digits that can indicate a best-before date, the format can specify that two numbers for the year, separated by a period or a slash, can be followed by two further numbers that can indicate the month. Figure 2b shows an image representing section 12 with the string 14 as image information 16.

[0042] The image information 16 can be received by means of an input signal according to step 102 of the computer-implemented method 100.

[0043] According to optional step 108, a region 18, 20 of particular interest can be defined in the image information 16 displaying section 12. As shown in Figure 2b, the string 14, for example, can have two lines, with the top line potentially displaying a best-before date and the bottom line potentially displaying a batch number. The top line can represent a first region 18. The second line can represent a second region 20.

[0044] As shown in Figures 2a and 2b, the concave shape of the can bottom 12 can distort the character string 14. Therefore, before determining the probability of the occurrence of the template character string 22, 24, a further optional step 110 can be performed to realign regions 18, 20 and / or the entire image information 16. This realignment aims to bring the character string 14 displayed in regions 18, 20 or the image information 16 into an orientation as similar as possible to the template character string, thus enabling the probability determination. This avoids a reduction in the determined probability caused by the orientation of the character string 14.

[0045] According to step 104, at least one evaluation value can be determined from the image information 16 using a trained, self-learning algorithm 26. The evaluation value indicates the minimum probability of the presence of the template string 22, 24 in the image information 16.

[0046] Figures 3a and 3b show example template strings 22 and 24. The template string 22 in Figure 3a could, for example, represent the sequence of digits for a best-before date. Two pairs of digits are separated by a period. The image information 16, or region 18, can be examined by the trained machine learning algorithm 26 using the template string 22 to calculate the probability of the presence of the template string 22, 24. Similarly, the probability of the presence of the template string 24 can be determined in region 20, where the template string 24 is a combination of letters and digits. The probability calculation can be based on the fact that, according to a template string format, letters are arranged at certain positions in the template string 22, 24, and digits at other positions.The specific letters can be defined, as can the specific characters, but this can also be left open. A mixture of open values ​​and specifically defined characters in different positions is also conceivable.

[0047] Figure 4 shows an example of a trained, self-learning algorithm 26, which can have an input level 28 and an output level 30. The trained, self-learning algorithm 26 can, for example, include an artificial neural network between the input level 28 and the output level 30.

[0048] The input level 28 can, for example, be configured to detect a part 34 of a region 18 or a part of the image information 16. As shown in Figure 5, the part 34 can, for example, scan the image information 16 or the region 18, whereby, for each detection of a part 34 by the input level 28, a probability value for a character can be provided at the output level 30. The output level 30 can be configured to output a probability value 32 for all characters that can be contained in the character string 14, indicating whether the corresponding character can be present in the part 34.

[0049] The explanations given above for input level 28 and output level 30 can apply analogously to the following examples.

[0050] According to a further embodiment, part 34 can first determine the position of the characters within region 18, 20 or the image information 16. Then, a subset of the image information 16 or of region 18, 20 can be provided to input level 28, containing only one character at a position in the character string 14. Output level 30 can then output the probability of each possible character being present at the corresponding position. This can be repeated for the other positions of the character string 14 that are detected.

[0051] In other embodiments, the region 18, 20, or the image information 16 in its entirety, can be captured by the input level. In these embodiments, the trained machine learning algorithm 26 can output a probability value for each of the possible characters at the output level 32 for each position in the character string 14.

[0052] The probabilities provided by output level 30 for the possible characters in the string 14 can be collected for all positions in a matrix 54, which is shown by way of example in Figure 6.

[0053] Each cell in the matrix represents a probability value for a specific character at a specific position in the string 14. In this example, the characters are digits and spaces. However, this does not preclude the possibility that the characters could also represent letters or other symbols.

[0054] Based on these probabilities, an assessment score can be determined, indicating the probability of the template string 22, 24 occurring in the region 18, 20. For example, for any string that can be generated by a combination of characters in different positions, a probability score can be determined by multiplying or otherwise combining the individual probabilities. Based on these determined probability scores, a score can then be calculated, which can be output as an assessment score. The assessment score can, for example, be in a range of 0-1 or 0-100. When determining the probabilities, it can also be taken into account if characters have a very low individual probability for a particular position.These characters can then be ignored for this position when analyzing the possible combinations of characters. A maximum allowed distance between individual characters in the string can also be specified, for example, by the template string format.

[0055] Furthermore, it is conceivable that the characters underlying the most probable character string could be output. This could be done in a separate output signal or together with the evaluation value.

[0056] Figure 7 shows a system 40 which can be designed for the production and / or treatment of objects 10, in particular containers.

[0057] The system 40 can include a device 50, 52 with which the objects 10 can be transported. The device 50, 52 can, for example, be designed for transporting objects 10, which may be shaped like cans. The device 50, 52 can include a vacuum transport mechanism 50 with which the cans can be held and transported at their head end. It is not excluded here that the objects 10 may be designed in other ways, for example as containers, in particular PET containers.

[0058] The vacuum transport system 50 can transport the objects 10 to a device 42, which is designed to apply at least one character string 14 to at least one section 12 of the objects 10. For example, laser encoding can be performed by the device 42, with the encoding being applied to the object as a character string 14. In this embodiment, the character string 14 can be applied to the bottom of the cans. However, this does not preclude the possibility that the character string 14 can also be applied to another position on the object 10.

[0059] In other embodiments, the device 50, 52 may not have vacuum transport, but rather another type of transport that can transport the objects 10, for example, to a labeling device. A labeling device can then label the objects 10 with labels that may contain a character string 14.

[0060] Alternatively or additionally, in other embodiments the character string 14 can be applied to the objects 10 by embossing or printing.

[0061] In the direction of transport, behind the device 42, a device for verifying the string 14 on the object 10 can be arranged. The device can have an inspection area 58 for the objects 10, which can be arranged on the device 50, 52. The objects 10 can be transported by the device 50, 52 into or through the inspection area 58.

[0062] Furthermore, the verification device can include an inspection device 44 that can inspect an object 10 in the inspection area 58. In this embodiment, the inspection device 44 is arranged such that it can inspect the base of the object 10 to which the string 14 has been attached. However, the inspection device 44 can also be arranged such that it can inspect other areas of the object.

[0063] Furthermore, the device for verification can include a processor unit 56, which can be configured to perform the method 100 described above. The processor unit 56 can be connected to the inspection device 44 for transmitting signals via a signal connection 60. The signal connection 60 can be wireless or wired.

[0064] In the direction of transport, downstream of the verification device, a device 46 for inspecting the interior of the objects 10 can be arranged. The device 46 can, for example, be positioned above the objects 10. The objects can be transported in the direction of transport by means of a conveyor belt 52. Objects 10 that have a character string 14 with a rating value below a predefined threshold can be rejected by a diverting device 48, which can be arranged downstream of the device 46 in the direction of transport. The threshold can be predefined such that rating values ​​that do not exceed the threshold indicate a low probability of the presence of the template character string 22, 24, for example, in region 18, 20.

[0065] In some embodiments, the discharge device 48 can also discharge objects 10 that have been classified as defective by the device 46.

[0066] The example described above in no way limits the invention. Rather, the invention can be modified in numerous ways. All features of the invention described above can be essential to the invention, either individually or in combination.

[0067] Reference symbol list

[0068] 10 objects

[0069] Section 12

[0070] 14 string

[0071] 16 Image Information

[0072] 18 Region

[0073] 20 Region

[0074] 22 Template string

[0075] 24 Template string

[0076] 26 Algorithm

[0077] 28 Input level

[0078] 30 Output level

[0079] 32 probability value

[0080] Part 34

[0081] 40 Annex

[0082] 42 Device for attaching at least one string

[0083] 44 Inspection equipment

[0084] 46 Device for inspecting an interior

[0085] 48 Exit device

[0086] 50 Device for transporting

[0087] 52 Device for transporting

[0088] 54 Matrix

[0089] 56 processor units

[0090] 58 Inspection area

[0091] 60 Signal connection

Claims

Claims 1. Computer-implemented method (100) for verifying at least one character string (14) attached to at least one object (10), in particular at least one container, which has characters at different positions, in particular an encoding, wherein the character string (14) is arranged on at least one section (12) of the object (10) and was produced according to at least one template character string (22, 24), comprising at least the following steps: Receiving (102) an input signal displaying image information (16) of at least section (12); Determine (104) at least one judgment value for the character string (14) displayed in the image information (16) using at least one trained machine learning algorithm (26), wherein the judgment value indicates at least one probability of the occurrence of the template character string (22, 24) in the image information (16); and Providing (106) at least one output signal indicating the at least one assessment value for verifying the string (14) on the object (10).

2. Computer-implemented method (100) according to claim 1, characterized in that the trained learning algorithm (26) comprises at least one artificial neural network, in particular a convolutional neural network, into which the image information (16) is input and which outputs the at least one evaluation value.

3. Computer-implemented method (100) according to claim 1 or 2, characterized in that the trained learning algorithm (26) outputs at least one matrix (54) of probabilities of finding a predefined character at a specific position in the string (14).

4. Computer-implemented method (100) according to one of the preceding claims, characterized in that, prior to determining the assessment value, at least one region (18, 20) of interest is defined (108) in the section (12) displayed in the image information (16), wherein the region (18, 20) has the string (14).

5. Computer-implemented method (100) according to claim 4, characterized in that in the step: Determining (104) at least one assessment value, only the image information (16) with which the region (18, 20) is displayed is used.

6. Computer-implemented method (100) according to claim 4 or 5, characterized in that, if several regions are defined, a separate assessment value is determined for each region (18, 20).

7. Computer-implemented method (100) according to one of the preceding claims, characterized in that the template string (22, 24) comprises a string with a predefined template string format, wherein the trained learning algorithm (26) determines a probability for the occurrence of a string (14) corresponding to the template string format.

8. Computer-implemented method (100) according to one of the preceding claims, characterized in that, prior to the step of determining (104) an evaluation value, at least one orientation of the image information (16) with the string (14) is determined (110) and at least partially adapted to at least one orientation of the template string (22, 24).

9. Computer-implemented method (100) according to one of the preceding claims, characterized in that a most probable string (14) is determined on the basis of the at least one evaluation value and is provided (112) as a further output signal.

10. Computer program product comprising instructions which, when the program is executed by a computer, cause it to perform the steps of the method (100) according to any of the preceding claims.

11. Device for verifying at least one string (14) attached to at least one object (10), in particular at least one container, wherein the string (14) is arranged on at least one section (12) of the object (10) and was produced according to at least one template string (22, 24), comprising at least one processor unit (56), an inspection device (44) and an inspection area (58) for at least one object (10) with at least one string (14) to be inspected, wherein the inspection device (44) is configured to determine image information (16) of at least one section (12) of an object (10) arranged in the inspection area (58) and to provide the image information (16), wherein the processor unit (56) is signal-connected to the inspection device (44) by means of a signal connection (60), characterized in thatthat the processor unit (56) is configured to perform the procedure (100) according to one of the steps 1 to 9.

12. System (40) for manufacturing and / or treating objects (10), comprising at least one device (50, 52) for transporting at least one object (10) to be treated along a transport direction, a device (42) for applying at least one string (14) in at least one section (12) of an object (10) which is arranged on the device (50, 52) for transporting, and at least one device according to claim 11 which is arranged in the transport direction behind the device (42) for applying a string (14) to the device (50, 52) for transporting, wherein the device (50, 52) for transporting extends through the inspection area (58).

Images