Electrode vision inspection device and method

Abstract

The present invention provides an electronic vision inspection device (1) comprising: a storage unit (100) which stores one or more pieces of training data, each of which includes input data including an image (G) obtained by imaging an electrode unit (70), and output data including electrode data related to one or more edges (E) or one or more corners (C) of each of one or more portions of the electrode unit (70) in the image (G); a model training unit (200) which machine-trains a first model so as to output the output data corresponding to input data input when the input data of each piece of the training data is input; an image acquisition unit (60) which acquires a first image (G1) obtained by imaging a first electrode unit (70A); a model application unit (300) which inputs first input data including the first image (G1) to the first model and outputs first output data including first electrode data corresponding to the electrode data; a position acquisition unit (400) which acquires or calculates, from the first electrode data, a first position of the one or more edges (E) or the one or more corners (C) of each of one or more portions of the first electrode unit (70A) in the first image (G1); and an evaluation unit (500) which determines, on the basis of the first position, whether the first electrode unit (70A) is defective.

Description

Electrode vision inspection device and method

[0001] This application claims the benefit of priority based on Korean Patent Application No. 10-2024-0139650 filed on October 14, 2024, and all contents disclosed in the document of said Korean patent application are incorporated herein as part of this specification.

[0002] The present invention relates to an electrode vision inspection device and method, wherein the detection capability of edges or corners of one or more parts of an electrode portion within an image is improved, non-detection, over-detection, and false detection are prevented, the accuracy of determining whether the electrode portion is defective is improved, and the detection capability of foreign matter located on the surface of the electrode portion is improved.

[0003] An electrode vision inspection device is used to detect defects in electrodes during manufacturing. The electrode vision inspection device determines whether an electrode is defective by analyzing images of the electrode acquired from cameras installed on process equipment.

[0004] Conventional electrode vision inspection devices detect characteristic edges or corners of multiple parts of an electrode using a rule-based method, and then determine whether the electrode is defective based on the location of the detected edges or corners. Here, the rule may, for example, detect pixels as edges or corners where the brightness difference with adjacent pixels within an image is greater than or equal to a reference value.

[0005] Therefore, when using an electrode vision inspection device, if shadows occur on the electrode due to wrinkles or bending, the detection capability of the edges or corners of each part of the electrode decreases, resulting in non-detection, over-detection, and false detection, which lowers the accuracy of determining whether the electrode is defective. In addition, it is difficult to detect foreign substances on the electrode surface using conventional rule-based methods. Therefore, a method to solve these problems is required.

[0006] Prior art related to this is Korean Published Patent No. 10-2024-0031887.

[0007] The present invention was devised to solve the aforementioned problems and aims to provide an electrode vision inspection device and method in which the detection capability of edges or corners of one or more parts of an electrode portion within an image is improved and non-detection, over-detection, and false detection are prevented.

[0008] In addition, the present invention aims to provide an electrode vision inspection device and method that improves the accuracy of determining whether an electrode part is defective.

[0009] In addition, the present invention aims to provide an electrode vision inspection device and method that are easily implemented at low cost.

[0010] In addition, the present invention aims to provide an electrode vision inspection device and method capable of easily generating training data and precisely retraining it.

[0011] In addition, the present invention aims to provide an electrode vision inspection device and method capable of easily and precisely verifying a machine-learned model.

[0012] In addition, the present invention aims to provide an electrode vision inspection device and method that allows the method for determining defects to be easily changed and the reliability of the determination to be easily adjusted.

[0013] In addition, the present invention aims to provide an electrode vision inspection device and method capable of automatically generating training data.

[0014] In addition, the present invention aims to provide an electrode vision inspection device and method capable of easily verifying judgment results.

[0015] In addition, the present invention aims to provide an electrode vision inspection device and method capable of detecting foreign matter located on the surface of an electrode and improving the detection capability of foreign matter.

[0016] In addition, the present invention aims to provide an electrode vision inspection device and method in which foreign matter on the surface of the electrode is distinguished from bubbles and correctly detected.

[0017] Furthermore, the present invention aims to provide an electrode vision inspection device and method that can machine learn a model quickly and effectively at low cost and is simple and easy to implement.

[0018] In addition, the present invention aims to provide an electrode vision inspection device and method capable of simply generating training data.

[0019] The technical problems of the present invention are not limited to the purposes mentioned above, and other unmentioned purposes and advantages of the present invention may be understood from the following description and will be more clearly understood by the embodiments of the present invention. Furthermore, it will be readily apparent that the purposes and advantages of the present invention can be realized by the means and combinations thereof set forth in the claims.

[0020] To solve the above-mentioned problem, the present invention provides an electrode vision inspection device (1) comprising a storage unit (100), a model learning unit (200), an image acquisition unit (60), a model application unit (300), a position acquisition unit (400), and an evaluation unit (500).

[0021] One or more training data can be stored in the storage unit (100) above.

[0022] Each of the above training data may include input data and output data.

[0023] The above input data may include an image (G) of the electrode part (70).

[0024] The above output data may include electrode data.

[0025] The electrode data may be data related to one or more edges (E) or one or more corners (C) of each of one or more parts of the electrode portion (70) in the image (G).

[0026] The above model learning unit (200) can machine learn the first model to output the output data corresponding to the input data when the input data of each of the above training data is input.

[0027] The above image acquisition unit (60) can acquire a first image (G1) of the first electrode unit (70A).

[0028] The above model application unit (300) can input first input data including the first image (G1) into the first model and output first output information data including first electrode data corresponding to the electrode data.

[0029] The above position acquisition unit (400) can acquire or calculate the first position of one or more edges (E) or one or more corners (C) of each of one or more parts of the first electrode part (70A) within the first image (G1) from the first electrode data.

[0030] The evaluation unit (500) can determine whether the first electrode unit (70A) is defective based on the first position.

[0031] In one embodiment, the first model may be a deep learning model.

[0032] In one embodiment, the electrode data may include position data of one or more edges (E) or one or more corners (C) of each of the one or more parts of the electrode portion (70) in the image (G).

[0033] The first electrode data may include first position data regarding the position of one or more edges (E) or one or more corners (C) of each of the one or more parts of the first electrode part (70A) within the first image (G1).

[0034] The above position acquisition unit (400) can acquire the first position from the above first position data.

[0035] In one embodiment, the output data may include segmentation data in which the image (G) is partitioned into a plurality of overlapping regions (R), each comprising one or more regions (R) corresponding to one or more parts of the electrode portion (70).

[0036] The electrode data may include at least a portion of the segmentation data.

[0037] The first output data above may include first segmentation data corresponding to the segmentation data above.

[0038] The first electrode data may include at least a portion of the first segmentation data.

[0039] The above position acquisition unit (400) can calculate the first position based on at least a portion of the first segmentation data included in the first electrode data.

[0040] In one embodiment, each of the electrode portion (70) and the first electrode portion (70A) may include an electrode sheet (72) and an electrode tab (76) that can be coupled to the electrode sheet (72).

[0041] Each of the above electrode portion (70) and the first electrode portion (70A) may include one or more portions of the electrode sheet (72) and the electrode tab (76).

[0042] The above position acquisition unit (400) can calculate the first position of one or more edges (E) or one or more corners (C) of each of at least one part of the electrode sheet (72) and the electrode tab (76) based on at least a portion of the first segmentation data included in the first electrode data.

[0043] In one embodiment, the electrode sheet (72) may include a retaining portion (A1) on which an electrode active material (74) is applied to an electrode foil (73) and an uncoated portion (A2) on which the electrode active material (74) is not applied to the electrode foil (73).

[0044] The above electrode tab (76) can be coupled to the above unpaired portion (A2).

[0045] Each of the above electrode portion (70) and the first electrode portion (70A) may include one or more portions of the above electrode portion (70) and the above electrode tab (76).

[0046] The above position acquisition unit (400) can calculate the first position of one or more edges (E) or one or more corners (C) of each of at least one part of the above-mentioned portion (A2) and the electrode tab (76) based on at least a portion of the above-mentioned first segmentation data included in the above-mentioned first electrode data.

[0047] In one embodiment, the position acquisition unit (400) can determine whether the electrode tab (76) exists based on at least a portion of the first electrode segmentation data included in the first electrode data.

[0048] The evaluation unit (500) can determine that the first electrode unit (70A) is defective when the electrode tab (76) is absent.

[0049] In one embodiment, each of the electrode portion (70) and the first electrode portion (70A) may further include a protective tape (78) that can be attached to the electrode sheet (72) or the electrode tab (76).

[0050] Each of the above electrode portion (70) and the first electrode portion (70A) may further include the above protective tape (78).

[0051] The above position acquisition unit (400) can calculate the first position of one or more edges (E) or one or more corners (C) of each of at least one part of the electrode sheet (72), electrode tab (76), and protective tape (78) based on at least a portion of the first segmentation data included in the first electrode data.

[0052] In one embodiment, the position acquisition unit (400) can determine whether the protective tape (78) is present based on at least a portion of the first segmentation data included in the first electrode data.

[0053] The evaluation unit (500) can determine that the first electrode unit (70A) is defective if the protective tape (78) is not present.

[0054] In one embodiment, the position acquisition unit (400) can calculate a second position of one or more edges (E) or one or more corners (C) of each of the one or more parts of the first electrode part (70A) in the first image (G1) by applying a predefined rule or method.

[0055] The evaluation unit (500) can calculate a first result determining whether the first electrode unit (70A) is defective based on the first position and a second result determining whether the first electrode unit (70A) is defective based on the second position, and can finally determine whether the first electrode unit (70A) is defective based on the first result and the second result.

[0056] In one embodiment, the method of the final determination may be selectable or set by the user.

[0057] In one embodiment, the method of the final determination may be selected or set as a first method or a second method.

[0058] The above first method may be a method for finally determining the first electrode part (70A) as defective when both the first result and the second result are defective.

[0059] The above second method may be a method for finally determining the first electrode part (70A) as defective when at least one of the above first result and second result is defective.

[0060] In one embodiment, when the first result is defective and the second result is normal, the model learning unit (200) can machine learn the first model to output the output data including the electrode data associated with one or more edges (E) or one or more corners (C) of each of the one or more parts of the first electrode unit (70A) corresponding to the second position when the input data including the first image (G1) is input.

[0061] In one embodiment, when the first result is normal and the second result is defective, the evaluation unit (500) can provide the first image (G1) to the user.

[0062] In one embodiment, one or more foreign substances (M) may be located on the surface of the electrode portion (70) and the first electrode portion (70A).

[0063] The output data of each of the above training data may include foreign material data related to one or more foreign materials (M) within the image (G).

[0064] The first output data above may include first foreign material data corresponding to the foreign material data above.

[0065] The above location acquisition unit (400) can determine whether there is one or more foreign substances (M) from the first foreign substance data, or calculate the third location, number, or area of ​​the one or more foreign substances (M).

[0066] The evaluation unit (500) may determine that the first electrode unit (70A) is defective when one or more foreign substances (M) are present, or determine whether the first electrode unit (70A) is defective based on the third location, number, or area of ​​the one or more foreign substances (M).

[0067] In one embodiment, the output data may include segmentation data in which the image (G) is partitioned into a plurality of overlapping regions (R), each comprising one or more regions (R4) corresponding to one or more foreign substances (M).

[0068] The above foreign matter data may include at least a portion of the above segmentation data.

[0069] The first output data above may include first segmentation data corresponding to the segmentation data above.

[0070] The first foreign material data above may include at least a portion of the first segmentation data.

[0071] The above location acquisition unit (400) can determine whether there is one or more foreign substances (M) based on at least a portion of the first segmentation data included in the first foreign substance data, or calculate the third location, number, or area.

[0072] In one embodiment, the segmentation data may be data obtained by partitioning the image (G) into a plurality of overlapping regions (R), each comprising one or more regions (R) corresponding to one or more parts of the electrode portion (70) and one or more regions (R4) corresponding to one or more foreign substances (M).

[0073] The electrode data may include at least a portion of the segmentation data.

[0074] The above position acquisition unit (400) can calculate the first position based on at least a portion of the first segmentation data included in the first electrode data.

[0075] In addition, to solve the above-mentioned problem, the present invention provides an electrode vision inspection device (1) comprising a storage unit (100), a model learning unit (200), a model application unit (300), a position acquisition unit (400), and an evaluation unit (500).

[0076] One or more training data can be stored in the storage unit (100).

[0077] Each of the above training data may include input data and output data.

[0078] The above input data may include an image (G) of the electrode part (70).

[0079] The above output data may include foreign matter data.

[0080] The above foreign matter data may be data related to one or more foreign matter (M) that may be located on the surface of the electrode part (70) within the image (G).

[0081] The above model learning unit (200) can machine learn the first model to output the output data corresponding to the input data when the input data of the training data is input.

[0082] The above image acquisition unit (60) can acquire a first image (G1) of the first electrode unit (70A).

[0083] The above model application unit (300) can input first input data including the first image (G1) into the first model and output first output data including first foreign material data corresponding to the foreign material data.

[0084] The above-mentioned position acquisition unit (400) can determine whether there is one or more foreign substances (M) on the surface of the first electrode part (70A) within the first image (G1) from the first foreign substance data, or calculate the third position, number, or area of ​​one or more foreign substances (M) on the surface of the first electrode part (70A).

[0085] The evaluation unit (500) may determine that the first electrode unit (70A) is defective when one or more foreign substances (M) are present, or determine whether the first electrode unit (70A) is defective based on the third location, number, or area of ​​one or more foreign substances (M).

[0086] In one embodiment, the output data may include segmentation data in which the image (G) is partitioned into a plurality of overlapping regions (R), each comprising one or more regions (R4) corresponding to one or more foreign substances (M) that can be located on the surface of the electrode portion (70).

[0087] The above foreign matter data may include at least a portion of the above segmentation data.

[0088] The first output data above may include first segmentation data corresponding to the segmentation data above.

[0089] The first foreign material data above may include at least a portion of the first segmentation data.

[0090] The above position acquisition unit (400) can determine whether there is one or more foreign substances (M) on the surface of the first electrode unit (70A) based on at least a portion of the first segmentation data included in the first foreign substance data, or calculate the third position, number, or area.

[0091] In addition, to solve the above-mentioned problem, the present invention provides an electrode vision inspection method (S900) comprising a model learning process (S910), an image acquisition process (S920), a model application process (S930), a position acquisition process (S940), and an evaluation process (S950).

[0092] In the above model learning process (S910), the model learning unit (200) can machine learn the first model so that when the input data of each of the training data is input, the output data of each of the training data corresponding to the input data is output.

[0093] The above input data may include the image (G) of the electrode part (70).

[0094] The output data above may include electrode data associated with one or more edges (E) or one or more corners (C) of each of one or more parts of the electrode portion (70) within the image (G).

[0095] In the above image acquisition process (S920), the image acquisition unit (60) can acquire the first image (G1) that captures the first electrode unit (70A).

[0096] In the above model application process (S930), the model application unit (300) can input the first input data including the first image (G1) into the first model and output the first electrode data corresponding to the electrode data.

[0097] In the above position acquisition process (S940), the position acquisition unit (400) may acquire or calculate the first position of one or more edges (E) or one or more corners (C) of each of the one or more parts of the first electrode part (70A) within the first image (G1) from the first electrode data.

[0098] In the above evaluation process (S950), the evaluation unit (500) can determine whether the first electrode unit (70A) is defective based on the first position.

[0099] In one embodiment, the electrode data may include position data regarding the location of one or more edges (E) or one or more corners (C) of each of the one or more parts of the electrode portion (70) in the image (G).

[0100] The first electrode data may include first position data regarding the position of one or more edges (E) or one or more corners (C) of each of the one or more parts of the first electrode part (70A) within the first image (G1).

[0101] In the above position acquisition process (S940), the position acquisition unit (400) can acquire the first position from the first position data.

[0102] In one embodiment, the output data may include segmentation data in which the image (G) is partitioned into a plurality of overlapping regions (R), each comprising one or more regions (R) corresponding to one or more parts of the electrode portion (70).

[0103] The electrode data may include at least a portion of the segmentation data.

[0104] The first output data above may include first segmentation data corresponding to the segmentation data above.

[0105] The first electrode data may include at least a portion of the first segmentation data.

[0106] In the above position acquisition process (S940), the position acquisition unit (400) can calculate the first position based on at least a portion of the first segmentation data included in the first electrode data.

[0107] In one embodiment, in the position acquisition process (S940), the position acquisition unit (400) may calculate a second position of an edge (E) or corner (C) of one or more parts of the first electrode part (70A) in the first image (G1) by applying a predefined rule or method.

[0108] In the above evaluation process (S950), the evaluation unit (500) can determine whether the first electrode unit (70A) is defective based on the first position and the first electrode unit (70A) is defective based on the second position, and can finally determine whether the first electrode unit (70A) is defective based on the first result and the second result.

[0109] In one embodiment, one or more foreign substances (M) may be located on the surface of the electrode portion (70) and the first electrode portion (70A).

[0110] The output data of each of the above training data may include foreign material data related to one or more foreign materials (M) within the image (G).

[0111] The first output data above may include first foreign material data corresponding to the foreign material data above.

[0112] In the above location acquisition process (S940), the location acquisition unit (400) can determine whether there is one or more foreign substances (M) from the first foreign substance data, or calculate the third location, number, or area of ​​the one or more foreign substances (M).

[0113] In the above evaluation process (S950), the evaluation unit (500) may determine that the first electrode unit (70A) is defective if one or more foreign substances (M) are present, or determine whether the first electrode unit (70A) is defective based on the third location, number, or area of ​​the one or more foreign substances (M).

[0114] In addition, to solve the above-mentioned problem, the present invention provides an electrode vision inspection method (S900) comprising a model learning process (S910), an image acquisition process (S920), a model application process (S930), a position acquisition process (S940), and an evaluation process (S950).

[0115] In the above model learning process (S910), the model learning unit (200) can machine learn the first model so that when the input data of each of the training data is input, the output data of each of the training data corresponding to the input data is output.

[0116] The above input data may include the image (G) of the electrode part (70).

[0117] The output data may include foreign matter data related to one or more foreign matter (M) that may be located on the surface of the electrode portion (70) within the image (G).

[0118] In the above image acquisition process (S920), the image acquisition unit (60) can acquire the first image (G1) that captures the first electrode unit (70A).

[0119] In the above model application process (S930), the model application unit (300) can input the first input data including the first image (G1) into the first model and output the first output data including the first foreign object data corresponding to the foreign object data.

[0120] In the above position acquisition process (S940), the position acquisition unit (400) can determine whether there is one or more foreign substances (M) on the surface of the first electrode unit (70A) within the first image (G1) from the first foreign substance data, or calculate the third position, number, or area of ​​the one or more foreign substances (M) on the surface of the first electrode unit (70A).

[0121] In the above evaluation process (S950), if one or more foreign substances (M) are present, the evaluation unit (500) may determine that the first electrode unit (70A) is defective, or determine whether the first electrode unit (70A) is defective based on the third location, number, or area of ​​the one or more foreign substances (M).

[0122] According to embodiments of the present invention, an electrode vision inspection device (1) comprises: a storage unit (100) in which one or more training data are stored, each including input data comprising an image (G) of an electrode part (70) and output data comprising electrode data related to one or more edges (E) or one or more corners (C) of each of one or more parts of the electrode part (70) within the image (G); a model learning unit (200) that machine-learns a first model to output output data corresponding to the input data when the input data of each of the training data is input; an image acquisition unit (60) that acquires a first image (G1) of a first electrode part (70A); and a model application unit (300) that inputs first input data including the first image (G1) to the first model and outputs first output data including first electrode data corresponding to the electrode data. It may include a position acquisition unit (400) that acquires or calculates a first position of one or more edges (E) or one or more corners (C) of each of one or more parts of the first electrode part (70A) within the first image (G1) from the first electrode data; and an evaluation unit (500) that determines whether the first electrode part (70A) is defective based on the first position.

[0123] Accordingly, by using a machine-learned first model to detect the edges (E) or corners (C) of one or more parts of the first electrode part (70A) within the first image (G1), the detection capability of the edges (E) or corners (C) is improved, and non-detection, over-detection, and false detection can be prevented. Accordingly, the accuracy of the defect determination of the electrode vision inspection device (1), which determines whether the electrode part (70) is defective based on the location of the edges (E) or corners (C) of one or more parts of the electrode part (70) detected within the image (G), can be improved. In particular, the accuracy of the defect determination can be improved compared to when the electrode vision inspection device (1) does not use a machine-learned model, or when it uses a machine-learned model but uses a model trained to directly output a defect determination result from the input image (G).

[0124] In addition, the first model can be machine-learned accurately and precisely using training data that includes accurate electrode data. Accordingly, the accuracy of determining whether the electrode vision inspection device (1) is defective can be improved.

[0125] In addition, the first model can be easily and precisely verified by methods such as displaying the contents of the electrode data (e.g., edges (E) or corners (C)) on the image (G). Furthermore, if the first model outputs incorrect electrode data, the first model can be precisely retrained to output correct electrode data.

[0126] According to embodiments of the present invention, the first model may be a deep learning model.

[0127] Accordingly, the detection capability of one or more edges (E) or one or more corners (C) of each of one or more parts of the electrode portion (70) in the input image (G) is improved, and non-detection, over-detection, and false detection can be prevented.

[0128] According to embodiments of the present invention, the electrode data may include position data of one or more edges (E) or one or more corners (C) of each of the one or more parts of the electrode portion (70) in the image (G). The first electrode data may include first position data regarding the position of one or more edges (E) or one or more corners (C) of each of the one or more parts of the first electrode portion (70A) in the first image (G1). The position acquisition unit (400) may acquire the first position from the first position data.

[0129] Accordingly, since the first model is trained to directly output position data regarding the position of one or more edges (E) or one or more corners (C) of each of one or more parts of the first electrode part (70A), the accuracy of the position of each edge (E) or each corner (C) acquired by the position acquisition part (400) can be improved. Accordingly, the detection capability of each edge (E) or each corner (C) is improved, and non-detection, over-detection, and false detection can be prevented. In addition, since the first position data output from the first model can be easily set or converted to the first position, the electrode vision inspection device (1) can be easily implemented at a low cost.

[0130] In addition, the first model can be easily and precisely verified by methods such as displaying each edge (E) or each corner (C) of the location data on the image (G). Furthermore, if the first model outputs incorrect location data, the first model can be precisely retrained to output correct location data. Additionally, since the location of the correct edge (E) or corner (C) in the image (G) can be accurately and easily found visually, correct training data can be easily generated.

[0131] According to embodiments of the present invention, the output data may include segmentation data in which the image (G) is partitioned into a plurality of overlapping regions (R), each comprising one or more regions (R) corresponding to one or more parts of the electrode portion (70). The electrode data may include at least a portion of the segmentation data. The first output data may include first segmentation data corresponding to the segmentation data. The first electrode data may include at least a portion of the first segmentation data. The position acquisition unit (400) may calculate the first position based on at least a portion of the first segmentation data included in the first electrode data.

[0132] Accordingly, after the corresponding region (R) for each part of the first electrode part (70A) is identified (divided) from the first image (G1), the edge (E) or corner (C) is detected from the identified (divided) region (R), so the detection capability of the edge (E) or corner (C) is improved and non-detection, over-detection, and false detection can be prevented. In particular, the edge (E) or corner (C) can be accurately detected even if the position or shape of one or more parts of the electrode part (70) changes.

[0133] Accordingly, the accuracy of determining defects of an electrode vision inspection device (1) can be improved based on the position of an edge (E) or corner (C) of one or more parts of an electrode part (70) detected in an image (G). In particular, the accuracy of determining defects can be improved compared to when the electrode vision inspection device (1) does not use a machine-learned model, or when it uses a machine-learned model but is trained to directly output a defect determination result from an input image (G).

[0134] In addition, the first model can be machine-learned accurately and precisely using training data that includes accurate segmentation data. Accordingly, the accuracy of determining whether the electrode vision inspection device (1) is defective can be improved.

[0135] In addition, since the region (R) of the electrode part (70) in the image (G) can be accurately and easily identified visually, correct training data can be easily generated.

[0136] In addition, the first model can be easily and precisely verified by methods such as displaying each region (R) of the segmentation data on the image (G). Furthermore, if the first model outputs incorrect segmentation data, the first model can be precisely retrained to output correct segmentation data.

[0137] According to embodiments of the present invention, each of the electrode portion (70) and the first electrode portion (70A) may include an electrode sheet (72) and an electrode tab (76) that can be coupled to the electrode sheet (72). Each of the electrode portion (70) and the first electrode portion (70A) may include the electrode sheet (72) and the electrode tab (76). The position acquisition unit (400) may calculate the first position of one or more edges (E) or one or more corners (C) of each of at least one portion of the electrode sheet (72) and the electrode tab (76) based on at least a portion of the first segmentation data included in the first electrode data.

[0138] Accordingly, after the electrode sheet region (e.g., the unoccupied region (R1)) and the electrode tab region (R2) are identified (divided) from the first image (G1), an edge (E) or a corner (C) can be detected from the identified (divided) region (R). Accordingly, the location of the electrode sheet (72) and / or the electrode tab (76) (e.g., the location of the edge or corner) can be accurately detected, and based on this, the defect status of the first electrode part (70A) can be accurately determined.

[0139] According to embodiments of the present invention, the electrode sheet (72) may include a retaining portion (A1) on which an electrode active material (74) is applied to an electrode foil (73) and a non-retaining portion (A2) on which the electrode active material (74) is not applied to the electrode foil (73). The electrode tab (76) may be attachable to the non-retaining portion (A2). Each of the electrode portion (70) and the first electrode portion (70A) may include the non-retaining portion (A2) and the electrode tab (76). The position acquisition unit (400) may calculate the first position of one or more edges (E) or one or more corners (C) of each of at least one portion of the non-retaining portion (A2) and the electrode tab (76) based on at least a portion of the first segmentation data included in the first electrode data.

[0140] Accordingly, after the blank area (R1) and the electrode tab area (R2) are identified (divided) from the first image (G1), an edge (E) or a corner (C) can be detected from the identified (divided) area (R). Accordingly, the location of the blank area (A2) and / or the electrode tab (76) (e.g., the location of the edge or corner) can be accurately detected, and based on this, the defect status of the first electrode part (70A) can be accurately determined.

[0141] According to embodiments of the present invention, the position acquisition unit (400) can determine whether the electrode tab (76) exists based on at least a portion of the first electrode segmentation data included in the first electrode data. The evaluation unit (500) can determine the first electrode unit (70A) as defective if the electrode tab (76) is not present.

[0142] Accordingly, the presence or absence of the electrode tab (76) can be accurately determined, and if the electrode tab (76) is not present, the first electrode part (70A) can be accurately determined as defective.

[0143] According to embodiments of the present invention, each of the electrode portion (70) and the first electrode portion (70A) may further include a protective tape (78) that can be attached to the electrode sheet (72) or the electrode tab (76). Each of the electrode portion (70) and the first electrode portion (70A) may further include the protective tape (78). The position acquisition unit (400) may calculate the first position of one or more edges (E) or one or more corners (C) of each of at least one portion of the electrode sheet (72), the electrode tab (76), and the protective tape (78) based on at least a portion of the first segmentation data included in the first electrode data.

[0144] Accordingly, after the electrode sheet area (e.g., the unmarked area (R1)), the electrode tab area (R2), and the protective tape area (R3) are identified (divided) from the first image (G1), an edge (E) or a corner (C) can be detected from the identified (divided) area (R). Accordingly, the location (e.g., the edge or corner location) of the electrode sheet (72), the electrode tab (76), and / or the protective tape (78) can be accurately detected, and based on this, the defect status of the first electrode part (70A) can be accurately determined.

[0145] According to embodiments of the present invention, the position acquisition unit (400) can determine whether the protective tape (78) is present based on at least a portion of the first segmentation data included in the first electrode data. The evaluation unit (500) can determine the first electrode unit (70A) as defective if the protective tape (78) is not present.

[0146] Accordingly, the presence or absence of the protective tape (78) can be accurately determined, and if the protective tape (78) is not present, the first electrode part (70A) can be accurately determined as defective.

[0147] According to embodiments of the present invention, the position acquisition unit (400) may calculate a second position of one or more edges (E) or one or more corners (C) of each of the one or more parts of the first electrode part (70A) in the first image (G1) by applying a predefined rule or method. The evaluation unit (500) may calculate a first result determining whether the first electrode part (70A) is defective based on the first position and a second result determining whether the first electrode part (70A) is defective based on the second position, and may finally determine whether the first electrode part (70A) is defective based on the first result and the second result.

[0148] Accordingly, the final determination of whether the first electrode part (70A) is defective is made based on two determination results calculated using two types of edge (E) or corner (C) detection methods (rule-based and machine learning), so the accuracy of determining whether the first electrode part (70A) is defective can be improved.

[0149] According to embodiments of the present invention, the method of the final determination may be selectable or set by the user.

[0150] Accordingly, the final judgment method can be easily changed and the reliability of the final judgment can be easily adjusted.

[0151] According to embodiments of the present invention, the method of final determination may be selected or set as a first method or a second method. The first method may be a method of finally determining the first electrode part (70A) as defective when both the first result and the second result are defective. The second method may be a method of finally determining the first electrode part (70A) as defective when at least one of the first result and the second result is defective.

[0152] Accordingly, the reliability of the final judgment can be easily adjusted.

[0153] According to embodiments of the present invention, when the first result is defective and the second result is normal, the model learning unit (200) can machine learn the first model to output the output data including the electrode data associated with one or more edges (E) or one or more corners (C) of each of the one or more parts of the first electrode unit (70A) corresponding to the second position when the input data including the first image (G1) is input.

[0154] Accordingly, training data (output data) capable of training the first model can be acquired or produced using a predefined rule or method (rule-based method). Accordingly, the first model can be easily (re)trained. In addition, training data capable of training the first model can be automatically generated.

[0155] According to embodiments of the present invention, when the first result is normal and the second result is defective, the evaluation unit (500) can provide the first image (G1) to the user.

[0156] Accordingly, the judgment results can be easily verified.

[0157] According to embodiments of the present invention, one or more foreign substances (M) may be located on the surface of the electrode portion (70) and the first electrode portion (70A). The output data of each of the training data may include foreign substance data related to the one or more foreign substances (M) within the image (G). The first output data may include first foreign substance data corresponding to the foreign substance data. The position acquisition unit (400) may determine whether the one or more foreign substances (M) exist from the first foreign substance data, or calculate a third position, number, or area of ​​the one or more foreign substances (M). The evaluation unit (500) may determine the first electrode portion (70A) as defective if the one or more foreign substances (M) exist, or determine whether the first electrode portion (70A) is defective based on the third position, number, or area of ​​the one or more foreign substances (M).

[0158] Accordingly, since a machine-learned model is used to detect foreign matter (M) on the surface of the first electrode part (70A) in the first image (G1), the detection capability of the foreign matter (M) can be improved. Accordingly, the electrode vision inspection device (1) can accurately determine whether the electrode part (70) is defective based on the foreign matter (M) detected on the surface of the first electrode part (70A) in the first image (G1). In particular, the accuracy of determining whether the electrode vision inspection device (1) is defective can be improved compared to when the electrode vision inspection device (1) does not use a machine-learned model, or when a machine-learned model is used but a model trained to directly output a defect determination result from the input image (G) is used. In addition, even if a bubble (B) generated when attaching the protective tape (78) is present on the first electrode part (70A), the foreign matter (M) can be correctly detected by distinguishing it from the bubble (B) (Figs. 6, 7).

[0159] In addition, the model can be machine-learned accurately and precisely using training data that includes accurate foreign object data. Accordingly, the accuracy of determining whether the electrode vision inspection device (1) is defective can be improved.

[0160] In addition, the model can be easily and precisely verified by methods such as marking the foreign object (M) of the foreign object data on the image (G). Furthermore, if the model outputs incorrect foreign object data, the model can be precisely retrained to output correct foreign object data.

[0161] According to embodiments of the present invention, the output data may include segmentation data in which the image (G) is partitioned into a plurality of overlapping regions (R), each comprising one or more regions (R4) corresponding to one or more foreign substances (M). The foreign substance data may include at least a portion of the segmentation data. The first output data may include first segmentation data corresponding to the segmentation data. The first foreign substance data may include at least a portion of the first segmentation data. The position acquisition unit (400) may determine the presence of the one or more foreign substances (M) or calculate the third position, number, or area based on at least a portion of the first segmentation data included in the first foreign substance data.

[0162] Accordingly, the detection capability of the foreign substance (M) can be improved by using a machine-learned model to identify (divide) the region (R4) of the foreign substance (M) from the first image (G1). In particular, the foreign substance (M) can be accurately detected even if the location, shape, size, etc. of the foreign substance (M) change. In addition, even if a bubble (B) generated when attaching the protective tape (78) is present in the first electrode part (70A), the foreign substance (M) can be correctly detected by distinguishing it from the bubble (B) (Figs. 6, 7).

[0163] In addition, the model can be machine-learned accurately and precisely using training data that includes accurate segmentation data. Accordingly, the accuracy of determining whether the electrode vision inspection device (1) is defective can be improved.

[0164] In addition, since the foreign object region (R4) in the image (G) can be accurately and easily identified visually, correct training data can be easily generated.

[0165] In addition, the model can be easily and precisely verified by methods such as marking the region (R4) of the foreign substance (M) of the segmentation data on the image (G). Furthermore, if the model outputs incorrect segmentation data, the model can be precisely retrained to output correct segmentation data.

[0166] According to embodiments of the present invention, the segmentation data may be data obtained by partitioning the image (G) into a plurality of overlapping regions (R), each comprising one or more regions (R) corresponding to one or more parts of the electrode portion (70) and one or more regions (R4) corresponding to one or more foreign substances (M). The electrode data may include at least a portion of the segmentation data. The position acquisition unit (400) may calculate the first position based on at least a portion of the first segmentation data included in the first electrode data.

[0167] Accordingly, using a single segmentation data, it is possible to detect the edge (E) or corner (C) and foreign matter (M) of one or more parts of the electrode part (70). Accordingly, the first model can be machine-learned quickly and effectively at low cost, and the electrode vision inspection device (1) can be implemented simply and easily. In addition, training data can be generated simply and easily.

[0168] According to embodiments of the present invention, an electrode vision inspection device (1) comprises: a storage unit (100) in which one or more training data are stored, each including input data comprising an image (G) of an electrode part (70) and output data comprising one or more foreign matter (M) that may be located on the surface of the electrode part (70) within the image (G); a model learning unit (200) that machine-learns a first model to output the output data corresponding to the input data when the input data of the training data is input; an image acquisition unit (60) that acquires a first image (G1) of a first electrode part (70A); and a model application unit (300) that inputs first input data including the first image (G1) to the first model and outputs first output data including first foreign matter data corresponding to the foreign matter data. The apparatus may include a position acquisition unit (400) that determines whether there is one or more foreign substances (M) on the surface of the first electrode part (70A) within the first image (G1) from the first foreign substance data, or calculates a third position, number, or area of ​​one or more foreign substances (M) on the surface of the first electrode part (70A); and an evaluation unit (500) that determines the first electrode part (70A) as defective when one or more foreign substances (M) exist, or determines whether the first electrode part (70A) is defective based on the third position, number, or area of ​​one or more foreign substances (M).

[0169] Accordingly, since a machine-learned model is used to detect foreign matter (M) on the surface of the first electrode part (70A) in the first image (G1), the detection capability of the foreign matter (M) can be improved. Accordingly, the electrode vision inspection device (1) can accurately determine whether the electrode part (70) is defective based on the foreign matter (M) detected on the surface of the first electrode part (70A) in the first image (G1). In particular, the accuracy of determining whether the electrode vision inspection device (1) is defective can be improved compared to when the electrode vision inspection device (1) does not use a machine-learned model, or when a machine-learned model is used but a model trained to directly output a defect determination result from the input image (G) is used. In addition, even if a bubble (B) generated when attaching the protective tape (78) is present on the first electrode part (70A), the foreign matter (M) can be correctly detected by distinguishing it from the bubble (B) (Figs. 6, 7).

[0170] In addition, the model can be machine-learned accurately and precisely using training data that includes accurate foreign object data. Accordingly, the accuracy of determining whether the electrode vision inspection device (1) is defective can be improved.

[0171] In addition, the model can be easily and precisely verified by methods such as marking the foreign object (M) of the foreign object data on the image (G). Furthermore, if the model outputs incorrect foreign object data, the model can be precisely retrained to output correct foreign object data.

[0172] According to embodiments of the present invention, the output data may include segmentation data in which the image (G) is partitioned into a plurality of overlapping regions (R), each comprising one or more regions (R4) corresponding to one or more foreign substances (M) that can be located on the surface of the electrode portion (70). The foreign substance data may include at least a portion of the segmentation data. The first output data may include first segmentation data corresponding to the segmentation data. The first foreign substance data may include at least a portion of the first segmentation data. The position acquisition unit (400) may determine the presence of foreign substances (M) on the surface of the first electrode portion (70A) or calculate the third position, number, or area based on at least a portion of the first segmentation data included in the first foreign substance data.

[0173] Accordingly, the detection capability of the foreign substance (M) can be improved by using a machine-learned model to identify (divide) the region (R4) of the foreign substance (M) from the first image (G1). In particular, the foreign substance (M) can be accurately detected even if the location, shape, size, etc. of the foreign substance (M) change. In addition, even if a bubble (B) generated when attaching the protective tape (78) is present in the first electrode part (70A), the foreign substance (M) can be correctly detected by distinguishing it from the bubble (B) (Figs. 6, 7).

[0174] In addition, the model can be machine-learned accurately and precisely using training data that includes accurate segmentation data. Accordingly, the accuracy of determining whether the electrode vision inspection device (1) is defective can be improved.

[0175] In addition, since the foreign object region (R4) in the image (G) can be accurately and easily identified visually, correct training data can be easily generated.

[0176] In addition, the model can be easily and precisely verified by methods such as marking the region (R4) of the foreign substance (M) of the segmentation data on the image (G). Furthermore, if the model outputs incorrect segmentation data, the model can be precisely retrained to output correct segmentation data.

[0177] According to embodiments of the present invention, an electrode vision inspection method (S900) comprises: a model learning process (S910) in which the model learning unit (200) inputs the input data of each of the training data, which includes the image (G) of the electrode part (70), and outputs the output data of each of the training data, which includes the electrode data corresponding to the input data and one or more edges (E) or one or more corners (C) of each of one or more parts of the electrode part (70) within the image (G); an image acquisition process (S920) in which the image acquisition unit (60) acquires the first image (G1) of the first electrode part (70A); and a model application process (S930) in which the model application unit (300) inputs the first input data including the first image (G1) to the first model and outputs the first electrode data corresponding to the electrode data. The above-mentioned position acquisition unit (400) may include a position acquisition process (S940) in which the above-mentioned position acquisition unit (400) acquires or calculates the first position of one or more edges (E) or one or more corners (C) of each of the one or more parts of the first electrode part (70A) within the first image (G1) from the first electrode data; and an evaluation process (S950) in which the above-mentioned evaluation unit (500) determines whether the first electrode part (70A) is defective based on the first position.

[0178] Accordingly, since the edge (E) or corner (C) of one or more parts of the first electrode part (70A) in the first image (G1) is detected using a machine-learned first model, the detection capability of the edge (E) or corner (C) is improved, and non-detection, over-detection, and false detection can be prevented. Accordingly, the accuracy of determining defects in the electrode vision inspection method (S900), which determines whether the electrode part (70) is defective based on the location of the edge (E) or corner (C) of one or more parts of the electrode part (70) detected in the image (G), can be improved. In particular, the accuracy of determining defects in the electrode vision inspection method (S900) can be improved compared to when the electrode vision inspection method (S900) does not use a machine-learned model, or when a machine-learned model is used but a model trained to directly output a defect determination result from the input image (G) is used.

[0179] In addition, the first model can be machine-learned accurately and precisely using training data that includes accurate electrode data. Accordingly, the accuracy of determining defects in the electrode vision inspection method (S900) can be improved.

[0180] In addition, the first model can be easily and precisely verified by methods such as displaying the contents of the electrode data (e.g., edges (E) or corners (C)) on the image (G). Furthermore, if the first model outputs incorrect electrode data, the first model can be precisely retrained to output correct electrode data.

[0181] According to embodiments of the present invention, the electrode data may include position data regarding the location of one or more edges (E) or one or more corners (C) of each of the one or more parts of the electrode portion (70) in the image (G). The first electrode data may include first position data regarding the location of one or more edges (E) or one or more corners (C) of each of the one or more parts of the first electrode portion (70A) in the first image (G1). In the position acquisition process (S940), the position acquisition unit (400) may acquire the first position from the first position data.

[0182] Accordingly, since the first model is trained to directly output position data regarding the position of one or more edges (E) or one or more corners (C) of each of one or more parts of the first electrode part (70A), the accuracy of the position of each edge (E) or each corner (C) acquired by the position acquisition part (400) can be improved. Accordingly, the detection capability of each edge (E) or each corner (C) is improved, and non-detection, over-detection, and false detection can be prevented. In addition, since the first position data output from the first model can be easily set or converted to the first position, the electrode vision inspection method (S900) can be easily implemented at low cost.

[0183] In addition, the first model can be easily and precisely verified by methods such as displaying each edge (E) or each corner (C) of the location data on the image (G). Furthermore, if the first model outputs incorrect location data, the first model can be precisely retrained to output correct location data. Additionally, since the location of the correct edge (E) or corner (C) in the image (G) can be accurately and easily found visually, correct training data can be easily generated.

[0184] According to embodiments of the present invention, the output data may include segmentation data in which the image (G) is partitioned into a plurality of overlapping regions (R), each comprising one or more regions (R) corresponding to one or more parts of the electrode portion (70). The electrode data may include at least a portion of the segmentation data. The first output data may include first segmentation data corresponding to the segmentation data. The first electrode data may include at least a portion of the first segmentation data. In the position acquisition process (S940), the position acquisition unit (400) may calculate the first position based on at least a portion of the first segmentation data included in the first electrode data.

[0185] Accordingly, after the corresponding region (R) for each part of the first electrode part (70A) is identified (divided) from the first image (G1), the edge (E) or corner (C) is detected from the identified (divided) region (R), so the detection capability of the edge (E) or corner (C) is improved and non-detection, over-detection, and false detection can be prevented. In particular, the edge (E) or corner (C) can be accurately detected even if the position or shape of one or more parts of the electrode part (70) changes.

[0186] In addition, the first model can be machine-learned accurately and precisely using training data that includes accurate segmentation data. Accordingly, the accuracy of determining defects in the electrode vision inspection method (S900) can be improved.

[0187] In addition, since the region (R) of the electrode part (70) in the image (G) can be accurately and easily identified visually, correct training data can be easily generated.

[0188] In addition, the first model can be easily and precisely verified by methods such as displaying each region (R) of the segmentation data on the image (G). Furthermore, if the first model outputs incorrect segmentation data, the first model can be precisely retrained to output correct segmentation data.

[0189] According to embodiments of the present invention, in the position acquisition process (S940), the position acquisition unit (400) may calculate a second position of an edge (E) or corner (C) of one or more parts of the first electrode part (70A) within the first image (G1) by applying a predefined rule or method. In the evaluation process (S950), the evaluation unit (500) may calculate a first result determining whether the first electrode part (70A) is defective based on the first position and a second result determining whether the first electrode part (70A) is defective based on the second position, and may finally determine whether the first electrode part (70A) is defective based on the first result and the second result.

[0190] Accordingly, the final determination of whether the first electrode part (70A) is defective is made based on two determination results calculated using two types of edge (E) or corner (C) detection methods (rule-based and machine learning), so the accuracy of determining whether the first electrode part (70A) is defective can be improved.

[0191] According to embodiments of the present invention, one or more foreign substances (M) may be located on the surface of the electrode portion (70) and the first electrode portion (70A). The output data of each of the training data may include foreign substance data related to the one or more foreign substances (M) within the image (G). The first output data may include first foreign substance data corresponding to the foreign substance data. In the position acquisition process (S940), the position acquisition unit (400) may determine the presence of the one or more foreign substances (M) from the first foreign substance data, or calculate a third position, number, or area of ​​the one or more foreign substances (M). In the above evaluation process (S950), the evaluation unit (500) may determine that the first electrode unit (70A) is defective if one or more foreign substances (M) are present, or determine whether the first electrode unit (70A) is defective based on the third location, number, or area of ​​the one or more foreign substances (M).

[0192] Accordingly, since a machine-learned model is used to detect foreign matter (M) on the surface of the electrode part (70) in the image (G), the detection capability of the foreign matter (M) can be improved. Accordingly, the electrode vision inspection method (S900) can accurately determine whether the electrode part (70) is defective based on the foreign matter (M) detected on the surface of the electrode part (70) in the image (G). In particular, the accuracy of determining whether the electrode part is defective can be improved compared to when the electrode vision inspection method (S900) does not use a machine-learned model, or when a machine-learned model is used but a model trained to directly output a result of determining whether the defect is defective from the input image (G) is used. In addition, even if bubbles (B) generated when attaching the protective tape (78) are present on the electrode part (70), the foreign matter (M) can be correctly detected by distinguishing it from the bubbles (B) (Figs. 6, 7).

[0193] In addition, the model can be machine-learned accurately and precisely using training data that includes accurate foreign object data. Accordingly, the accuracy of determining defects in the electrode vision inspection method (S900) can be improved.

[0194] In addition, the model can be easily and precisely verified by methods such as displaying foreign object data on the image (G). Furthermore, if the model outputs incorrect foreign object data, the model can be precisely retrained to output correct foreign object data.

[0195] According to embodiments of the present invention, an electrode vision inspection method (S900) comprises: a model learning process (S910) in which the model learning unit (200) inputs the input data of each training data, which includes the image (G) of the electrode part (70), and outputs the output data of each training data, which includes the foreign matter data corresponding to the input data and which may be located on the surface of the electrode part (70) within the image (G); an image acquisition process (S920) in which the image acquisition unit (60) acquires the first image (G1) of the first electrode part (70A); and a model application process (S930) in which the model application unit (300) inputs the first input data including the first image (G1) to the first model and outputs the first output data including the first foreign matter data corresponding to the foreign matter data. The above-mentioned position acquisition unit (400) may include a position acquisition process (S940) in which the above-mentioned position acquisition unit (400) determines whether there is one or more foreign substances (M) on the surface of the first electrode part (70A) within the first image (G1) from the first foreign substance data, or calculates a third position, number, or area of ​​the one or more foreign substances (M) on the surface of the first electrode part (70A); and an evaluation process (S950) in which the above-mentioned evaluation unit (500) determines the first electrode part (70A) as defective when the one or more foreign substances (M) exist, or determines whether the first electrode part (70A) is defective based on the third position, number, or area of ​​the one or more foreign substances (M).

[0196] Accordingly, since a machine-learned model is used to detect foreign matter (M) on the surface of the first electrode part (70A) in the first image (G1), the detection capability of the foreign matter (M) can be improved. Accordingly, the electrode vision inspection method (S900) can accurately determine whether the electrode part (70) is defective based on the foreign matter (M) detected on the surface of the first electrode part (70A) in the first image (G1). In particular, the accuracy of determining defects can be improved compared to when the electrode vision inspection method (S900) does not use a machine-learned model, or when a machine-learned model is used but a model trained to directly output a defect determination result from the input image (G) is used. In addition, even if bubbles (B) generated when attaching the protective tape (78) are present on the first electrode part (70A), the foreign matter (M) can be correctly detected by distinguishing it from the bubbles (B) (Figs. 6, 7).

[0197] In addition, the model can be machine-learned accurately and precisely using training data that includes accurate foreign object data. Accordingly, the accuracy of determining defects in the electrode vision inspection method (S900) can be improved.

[0198] In addition, the model can be easily and precisely verified by methods such as marking the foreign object (M) of the foreign object data on the image (G). Furthermore, if the model outputs incorrect foreign object data, the model can be precisely retrained to output correct foreign object data.

[0199] In addition to the effects described above, the specific effects of the present invention are described together with the specific details for implementing the invention below.

[0200] FIG. 1 is a schematic diagram showing an electrode vision inspection device according to one embodiment of the present invention.

[0201] Figure 2 is a block diagram of the operation unit of Figure 1.

[0202] FIG. 3 is an example of an image acquired by the image acquisition unit of FIG. 1.

[0203] Figure 4 is a drawing in which characteristic edges and corners of the image in Figure 3 are indicated by dotted lines and dots, respectively.

[0204] Figure 5 is a drawing in which three regions corresponding to three parts of the electrode portion in the image of Figure 3 are indicated by dotted and dashed lines.

[0205] FIG. 6 is another embodiment of an image obtained from the image acquisition unit of FIG. 1.

[0206] Figure 7 is a drawing in which three regions corresponding to three parts of the electrode portion and a foreign matter region are indicated by dotted and dashed lines in the image of Figure 6.

[0207] FIG. 8 is a flowchart of an electrode vision inspection method according to one embodiment of the present invention.

[0208] FIG. 9 is an embodiment that specifies the position acquisition and evaluation processes of FIG. 8.

[0209] [Explanation of the symbol]

[0210] 1: Electrode vision inspection device

[0211] 10: Computation Unit 60: Image Acquisition Unit

[0212] 70: Electrode section 70A: First electrode section

[0213] 72: Electrode sheet

[0214] 73: Electrode foil 74: Electrode active material

[0215] 76: Electrode tab 78: Protective tape

[0216] A1: Maintenance Department A2: Non-maintenance Department

[0217] 100: Storage unit 200: Model training unit

[0218] 300: Model application unit 400: Position acquisition unit

[0219] 500: Evaluation Department

[0220] G: Image G1: First image

[0221] C: Corner E: Edge

[0222] R: Area

[0223] R1: Non-existent area R2: Electrode tab area

[0224] R3: Protective tape area R4: Foreign matter area

[0225] M: Foreign matter B: Air bubbles

[0226] The aforementioned objectives, features, and advantages are described in detail below with reference to the attached drawings, thereby enabling those skilled in the art to easily implement the technical concept of the present invention. In describing the present invention, detailed descriptions of known technologies related to the present invention are omitted if it is determined that such descriptions would unnecessarily obscure the essence of the invention. Hereinafter, preferred embodiments according to the present invention will be described in detail with reference to the attached drawings. In the drawings, the same reference numerals are used to indicate the same or similar components.

[0227] Although terms such as "first," "second," etc., are used to describe various components, it goes without saying that these components are not limited by these terms. These terms are used merely to distinguish one component from another, and unless specifically stated otherwise, the first component may also be the second component.

[0228] Throughout the specification, unless specifically stated otherwise, each component may be singular or plural.

[0229] In the following, the statement that any configuration is placed on the "upper (or lower)" of a component or on the "upper (or lower)" of a component may mean not only that any configuration is placed in contact with the upper (or lower) surface of said component, but also that another configuration may be interposed between said component and any configuration placed on (or below) said component.

[0230] In addition, where it is stated that one component is "connected," "combined," or "connected" to another component, it should be understood that while the components may be directly connected or connected to each other, another component may be "interposed" between each component, or each component may be "connected," "combined," or "connected" through another component.

[0231] Singular expressions used in this specification include plural expressions unless the context clearly indicates otherwise. In this application, terms such as "composed of" or "comprising" should not be interpreted as necessarily including all of the various components or steps described in the specification, and should be interpreted as meaning that some of the components or steps may be omitted or additional components or steps may be included.

[0232] FIG. 1 is a schematic diagram showing an electrode vision inspection device according to an embodiment of the present invention. FIG. 2 is a block diagram of the computation unit of FIG. 1. FIG. 3 is an example of an image acquired by the image acquisition unit of FIG. 1. FIG. 4 is a diagram showing characteristic edges and corners in the image of FIG. 3 indicated by dotted lines and dots, respectively. FIG. 5 is a diagram showing three regions corresponding to three parts of the electrode part in the image of FIG. 3 indicated by dotted and dashed lines. FIG. 6 is another example of an image acquired by the image acquisition unit of FIG. 1. FIG. 7 is a diagram showing three regions corresponding to three parts of the electrode part and a foreign matter region in the image of FIG. 6 indicated by dotted and dashed lines. FIG. 8 is a flowchart of an electrode vision inspection method according to an embodiment of the present invention. FIG. 9 is an example of a specific embodiment of the position acquisition process and evaluation process of FIG. 8.

[0233] [First embodiment of electrode vision inspection device]

[0234] Referring to FIGS. 1 and 2, the electrode vision inspection device (1) according to the first embodiment may include a computation unit (10) and an image acquisition unit (60). Let's first look at the image acquisition unit (60).

[0235] The image acquisition unit (60) may be a camera. The image acquisition unit (60) may acquire an image (G) of the electrode unit (70). The image acquisition unit (60) may acquire a first image (G1) of the first electrode unit (70A) (Figs. 3, 6).

[0236] Here, the electrode portion (70) and the first electrode portion (70A) may be the same or different from each other. The image (G) may be acquired before the first image (G1). This is because the first image (G1) is input to the first or second model learned using the image (G).

[0237] Additionally, each of the electrode portion (70) and the first electrode portion (70A) may include an electrode sheet (72) and an electrode tab (76) that can be coupled (e.g., welded) to the electrode sheet (72). Here, the electrode sheet (72) may include a retaining portion (A1) on which an electrode active material (74) is applied to an electrode foil (73) and a non-retaining portion (A2) on which the electrode active material (74) is not applied to the electrode foil (73). The electrode tab (76) may be coupled to the non-retaining portion (A2).

[0238] Additionally, each of the electrode portion (70) and the first electrode portion (70A) may further include a protective tape (78) that can be attached to the electrode sheet (72) or the electrode tab (76). The protective tape (78) can be attached to the electrode sheet (72) and the electrode tab (76) simultaneously. Accordingly, the protective tape (78) can strengthen the bonding force between the electrode sheet (72) (e.g., the uncoated portion (A2)) and the electrode tab (76). Additionally, by attaching the protective tape (78) to the electrode tab (76), the electrode tab (76) can be protected and electrically insulated.

[0239] In addition, one or more foreign substances (M) may be located on the surface of the electrode portion (70) and the first electrode portion (70A) (Fig. 6).

[0240] The computation unit (10) can be connected to the image acquisition unit (60). The computation unit (10) can receive an image (G) from the image acquisition unit (60). The computation unit (10) can receive a first image (G1) from the image acquisition unit (60).

[0241] The computation unit (10) may include a storage unit (100), a model learning unit (200), a model application unit (300), a position acquisition unit (400), and an evaluation unit (500). Below, we will examine each component of the computation unit (10).

[0242] [Storage Section]

[0243] One or more training data can be stored in the storage unit (100). Each training data may include input data and output data.

[0244] The input data may include an image (G). The image (G) can be obtained by photographing the electrode portion (70).

[0245] The output data may include electrode data. The output data may include foreign matter data. The output data may include segmentation data.

[0246] The electrode data may be data associated with one or more edges (E) or one or more corners (C) of each of one or more parts of the electrode portion (70) in the image (G) (Fig. 4).

[0247] Here, each part of the electrode portion (70) may be each component of the electrode portion (70) (electrode sheet (72), retaining portion (A1), non-retaining portion (A2), electrode tab (76), or protective tape (78)). However, it is not limited thereto. That is, each part of the electrode portion (70) may be a part of the electrode portion (70) unrelated to each component of the electrode portion (70). In addition, at least one of the above parts of the electrode portion (70) may be the entire electrode portion (70) within the image (G).

[0248] Additionally, one or more portions of the electrode portion (70) may include an electrode sheet (72) and an electrode tab (76). For example, one or more portions of the electrode portion (70) may include a blank portion (A2) and an electrode tab (76). One or more portions of the electrode portion (70) may further include a protective tape (78).

[0249] In one embodiment, the electrode data may include position data.

[0250] The position data may be data regarding the position of one or more edges (E) or one or more corners (C) of each of one or more parts of the electrode portion (70) in the image (G). For example, the position data may be data regarding the position of the edge (E) or the corner (C) of all or part (i.e., characteristic) of each of the electrode sheet (72) (e.g., the blank portion (A2)), electrode tab (76), and / or protective tape (78) in the image (G) (Fig. 4).

[0251] For example, the location data may include coordinate values ​​of each edge (E) or each corner (C). As another example, the location data may include a discrimination value corresponding to each pixel of the image (G). In this case, for each of one or more parts of the electrode portion (70), the discrimination value of one or more pixels corresponding to all or part of the edge (E) or corner (C) may be different from the discrimination value of other pixels.

[0252] In another embodiment, the electrode data may include at least a portion of the segmentation data described below. Specifically, the electrode data may include data of a portion related to one or more edges (E) or one or more corners (C) of each of one or more parts of the electrode portion (70) within the image (G) among the segmentation data described below.

[0253] The foreign material data may be data related to one or more of the aforementioned foreign materials (M) in the image (G) (Fig. 7).

[0254] In one embodiment, the foreign object data may include data regarding the location of one or more foreign objects (M). For example, the foreign object data may include coordinate values ​​of one or more foreign objects (M). As another example, the foreign object data may include a discrimination value corresponding to each pixel of the image (G). In this case, the discrimination value of one or more pixels corresponding to one or more foreign objects (M) may be different from the discrimination value of other pixels.

[0255] In another embodiment, the foreign object data may include at least a portion of the segmentation data described below. Specifically, the foreign object data may include data of a portion of the segmentation data described below that is related to one or more foreign objects (M) within the image (G).

[0256] Segmentation data may be data obtained by partitioning an image (G) into multiple overlapping regions (R) (Fig. 5). Here, the multiple regions (R) may include one or more regions (R) corresponding to one or more parts of the electrode portion (70). The multiple regions (R) may include one or more regions (R) corresponding to one or more of the aforementioned foreign substances (M). The multiple regions (R) may further include a background region.

[0257] For example, the segmentation data may be data divided into multiple overlapping regions (R) including three regions (R1, R2, R3) corresponding to the electrode sheet (72) (e.g., bare area (A2)), electrode tab (76), and protective tape (78) in the image (G), respectively, and a background (Fig. 5). As another example, the segmentation data may be data divided into multiple overlapping regions (R) including one or more regions (R4) corresponding to one or more foreign substances (M) in the image (G), respectively (Fig. 7).

[0258] For example, segmentation data may include label values ​​corresponding to each pixel of an image (G). Label values ​​may be set differently for each region (R). Accordingly, the label values ​​of pixels belonging to the same region (R) may be the same, while the label values ​​of pixels belonging to different regions (R) may be different. In the case where multiple regions (R) overlap, each pixel belonging to an overlapping region (R) may have multiple label values ​​corresponding to each of the overlapping regions (R).

[0259] Any matters not mentioned regarding segmentation may follow the announced technology regarding image segmentation using deep learning. The same applies below.

[0260] [Model Learning Section]

[0261] The model learning unit (200) can receive one or more training data from the storage unit (100).

[0262] The model learning unit (200) can machine learn the first model. Specifically, the model learning unit (200) can machine learn the first model so that when input data of each training data is input, it outputs output data of each training data corresponding to the input data.

[0263] Here, the first model may be a deep learning model. The deep learning model may be an artificial neural network model comprising an input layer, an output layer, and a plurality of hidden layers between the input layer and the output layer.

[0264] Accordingly, the detection capability of one or more edges (E) or one or more corners (C) of each of one or more parts of the electrode portion (70) in the input image (G) is improved, and non-detection, over-detection, and false detection can be prevented.

[0265] [Model Application Section]

[0266] The model application unit (300) can input first input data into a first model machine-learned with one or more training data to output first output data.

[0267] The first input data may include a first image (G1).

[0268] The first output data may include first electrode data corresponding to the aforementioned electrode data. The first output data may include first foreign matter data corresponding to the aforementioned foreign matter data. The first output data may include first segmentation data corresponding to the aforementioned segmentation data.

[0269] The first electrode data, first foreign matter data, and first segmentation data are as follows.

[0270] In one embodiment, the first electrode data may include first position data. The first position data may be data regarding the position of one or more edges (E) or one or more corners (C) of each of one or more parts of the first electrode portion (70A) in the first image (G1). For example, the first position data may be data regarding the position of the edge (E) or the corner (C) of all or part (i.e., characteristic) of each of the electrode sheet (72) (e.g., the blank portion (A2)), electrode tab (76), and / or protective tape (78) in the first image (G1) (Fig. 4). The first position data may have a data structure corresponding to the aforementioned position data and may be inferred from the aforementioned position data.

[0271] In another embodiment, the first electrode data may include at least a portion of the first segmentation data described below. Specifically, the first electrode data may include data of a portion related to one or more edges (E) or one or more corners (C) of each of one or more parts of the first electrode portion (70A) within the first image (G1) among the first segmentation data described below.

[0272] The first foreign object data may be data related to one or more of the aforementioned foreign objects (M) in the first image (G1) (Fig. 7).

[0273] In one embodiment, the first foreign object data may include data regarding the location of one or more foreign objects (M) as described above.

[0274] In another embodiment, the first foreign object data may include at least a portion of the first segmentation data described below. Specifically, the first foreign object data may include data of a portion of the first segmentation data described below that is related to one or more foreign objects (M) within the first image (G1).

[0275] The first segmentation data may be data obtained by partitioning the first image (G1) into multiple overlapping regions (R) (Fig. 5). Here, the multiple regions (R) may include one or more regions (R) corresponding to one or more parts of the first electrode portion (70A). The multiple regions (R) may include one or more regions (R4) corresponding to one or more foreign substances (M). The multiple regions (R) may further include a background region.

[0276] For example, the first segmentation data may be data divided into a plurality of overlapping regions (R) including three regions (R1, R2, R3) and a background region corresponding to the electrode sheet (72) (e.g., bare area (A2)), electrode tab (76), and protective tape (78) respectively within the first image (G1) (Fig. 5). As another example, the first segmentation data may be data divided into a plurality of overlapping regions (R) including one or more regions (R4) corresponding to one or more foreign substances (M) respectively within the first image (G1) (Fig. 7).

[0277] The first segmentation data may have a data structure corresponding to the aforementioned segmentation data and can be inferred from the segmentation data.

[0278] [Location Acquisition Section]

[0279] The position acquisition unit (400) can acquire or calculate the first position of one or more edges (E) or one or more corners (C) of each of one or more parts of the first electrode part (70A) in the first image (G1) from the first electrode data.

[0280] Here, one or more portions of the first electrode portion (70A) may include an electrode sheet (72) and an electrode tab (76). One or more portions of the first electrode portion (70A) may include a blank portion (A2) and an electrode tab (76). One or more portions of the first electrode portion (70A) may further include a protective tape (78).

[0281] In one embodiment, the position acquisition unit (400) can acquire the first position from the aforementioned first position data of the first electrode data. For example, the position acquisition unit (400) can acquire the first position of the edge (E) or corner (C) of all or part (i.e., characteristic) of each of one or more parts of the first electrode part (70A) in the first image (G1) from the first electrode data (Fig. 4).

[0282] Accordingly, since the first model is trained to directly output position data regarding the position of one or more edges (E) or one or more corners (C) of each of one or more parts of the first electrode part (70A), the accuracy of the position of each edge (E) or each corner (C) acquired by the position acquisition part (400) can be improved. Accordingly, the detection capability of each edge (E) or each corner (C) is improved, and non-detection, over-detection, and false detection can be prevented. In addition, since the first position data output from the first model can be easily set or converted to the first position, the electrode vision inspection device (1) can be easily implemented at a low cost.

[0283] In addition, the first model can be easily and precisely verified by methods such as displaying each edge (E) or each corner (C) of the location data on the image (G). Furthermore, if the first model outputs incorrect location data, the first model can be precisely retrained to output correct location data. Additionally, since the location of the correct edge (E) or corner (C) in the image (G) can be accurately and easily found visually, correct training data can be easily generated.

[0284] In another embodiment, the position acquisition unit (400) may calculate the first position based on at least a portion of the first segmentation data included in the first electrode data. For example, the position acquisition unit (400) may calculate the first position of an edge (E) or a corner (C) from the edge of each region (R) of the first segmentation data.

[0285] Accordingly, after the corresponding region (R) for each part of the first electrode part (70A) is identified (divided) from the first image (G1), the edge (E) or corner (C) is detected from the identified (divided) region (R), so the detection capability of the edge (E) or corner (C) is improved and non-detection, over-detection, and false detection can be prevented. In particular, the edge (E) or corner (C) can be accurately detected even if the position or shape of one or more parts of the electrode part (70) changes.

[0286] Accordingly, the accuracy of determining defects of an electrode vision inspection device (1) can be improved based on the position of an edge (E) or corner (C) of one or more parts of an electrode part (70) detected in an image (G). In particular, the accuracy of determining defects can be improved compared to when the electrode vision inspection device (1) does not use a machine-learned model, or when it uses a machine-learned model but is trained to directly output a defect determination result from an input image (G).

[0287] In addition, the first model can be machine-learned accurately and precisely using training data that includes accurate segmentation data. Accordingly, the accuracy of determining whether the electrode vision inspection device (1) is defective can be improved.

[0288] In addition, since the region (R) of the electrode part (70) in the image (G) can be accurately and easily identified visually, correct training data can be easily generated.

[0289] In addition, the first model can be easily and precisely verified by methods such as displaying each region (R) of the segmentation data on the image (G). Furthermore, if the first model outputs incorrect segmentation data, the first model can be precisely retrained to output correct segmentation data.

[0290] The position acquisition unit (400) can calculate the first position of at least one edge (E) or corner (C) of the electrode sheet (72) and the electrode tab (76) based on at least a portion of the first segmentation data included in the first electrode data.

[0291] Accordingly, after the electrode sheet region (e.g., the unoccupied region (R1)) and the electrode tab region (R2) are identified (divided) from the first image (G1), an edge (E) or a corner (C) can be detected from the identified (divided) region (R). Accordingly, the location of the electrode sheet (72) and / or the electrode tab (76) (e.g., the location of the edge or corner) can be accurately detected, and based on this, the defect status of the first electrode part (70A) can be accurately determined.

[0292] The position acquisition unit (400) can calculate the first position of one or more edges (E) or one or more corners (C) of each of at least one part of the unoccupied part (A2) and the electrode tab (76) based on at least a portion of the first segmentation data included in the first electrode data.

[0293] Accordingly, after the blank area (R1) and the electrode tab area (R2) are identified (divided) from the first image (G1), an edge (E) or a corner (C) can be detected from the identified (divided) area (R). Accordingly, the location of the blank area (A2) and / or the electrode tab (76) (e.g., the location of the edge or corner) can be accurately detected, and based on this, the defect status of the first electrode part (70A) can be accurately determined.

[0294] The position acquisition unit (400) can determine whether the electrode tab (76) exists based on at least a portion of the first segmentation data included in the first electrode data. For example, the position acquisition unit (400) can determine that the electrode tab (76) does not exist if there is no pixel having a label value corresponding to the electrode tab area (R2) in the first segmentation data.

[0295] Accordingly, the presence or absence of the electrode tab (76) can be accurately determined, and if the electrode tab (76) is not present, the first electrode part (70A) can be accurately determined as defective.

[0296] The position acquisition unit (400) can calculate a first position of one or more edges (E) or one or more corners (C) of each of at least one part of the electrode sheet (72) (e.g., unmarked part (A2)), electrode tab (76), and protective tape (78) based on at least a portion of the first segmentation data included in the first electrode data.

[0297] Accordingly, after the electrode sheet area (e.g., the unmarked area (R1)), the electrode tab area (R2), and the protective tape area (R3) are identified (divided) from the first image (G1), an edge (E) or a corner (C) can be detected from the identified (divided) area (R). Accordingly, the location (e.g., the edge or corner location) of the electrode sheet (72), the electrode tab (76), and / or the protective tape (78) can be accurately detected, and based on this, the defect status of the first electrode part (70A) can be accurately determined.

[0298] The position acquisition unit (400) can determine whether the protective tape (78) exists based on at least a portion of the first segmentation data included in the first electrode data. For example, the position acquisition unit (400) can determine that the protective tape (78) does not exist if there is no pixel having a label value corresponding to the protective tape area (R3) in the first segmentation data.

[0299] Accordingly, the presence or absence of the protective tape (78) can be accurately determined, and if the protective tape (78) is not present, the first electrode part (70A) can be accurately determined as defective.

[0300] The location acquisition unit (400) can determine whether there is one or more foreign substances (M) from the aforementioned first foreign substance data, or calculate the third location, number, or area of ​​one or more foreign substances (M).

[0301] Accordingly, since a machine-learned model is used to detect foreign matter (M) on the surface of the first electrode part (70A) in the first image (G1), the detection capability of the foreign matter (M) can be improved. Accordingly, the electrode vision inspection device (1) can accurately determine whether the electrode part (70) is defective based on the foreign matter (M) detected on the surface of the first electrode part (70A) in the first image (G1). In particular, the accuracy of determining whether the electrode vision inspection device (1) is defective can be improved compared to when the electrode vision inspection device (1) does not use a machine-learned model, or when a machine-learned model is used but a model trained to directly output a defect determination result from the input image (G) is used. In addition, even if a bubble (B) generated when attaching the protective tape (78) is present on the first electrode part (70A), the foreign matter (M) can be correctly detected by distinguishing it from the bubble (B) (Figs. 6, 7).

[0302] In addition, the model can be machine-learned accurately and precisely using training data that includes accurate foreign object data. Accordingly, the accuracy of determining whether the electrode vision inspection device (1) is defective can be improved.

[0303] In addition, the model can be easily and precisely verified by methods such as marking the foreign object (M) of the foreign object data on the image (G). Furthermore, if the model outputs incorrect foreign object data, the model can be precisely retrained to output correct foreign object data.

[0304] The location acquisition unit (400) can determine whether there is one or more of the aforementioned foreign substances (M) based on at least a portion of the first segmentation data included in the first foreign substance data, or calculate the third location, number, or area. For example, the location acquisition unit (400) can determine that the foreign substance (M) exists if there is a pixel having a label value corresponding to the foreign substance area (R4) in the first segmentation data.

[0305] Accordingly, the detection capability of the foreign substance (M) can be improved by using a machine-learned model to identify (divide) the region (R4) of the foreign substance (M) from the first image (G1). In particular, the foreign substance (M) can be accurately detected even if the location, shape, size, etc. of the foreign substance (M) change. In addition, even if a bubble (B) generated when attaching the protective tape (78) is present in the first electrode part (70A), the foreign substance (M) can be correctly detected by distinguishing it from the bubble (B) (Figs. 6, 7).

[0306] In addition, the model can be machine-learned accurately and precisely using training data that includes accurate segmentation data. Accordingly, the accuracy of determining whether the electrode vision inspection device (1) is defective can be improved.

[0307] In addition, since the foreign object region (R4) in the image (G) can be accurately and easily identified visually, correct training data can be easily generated.

[0308] In addition, the model can be easily and precisely verified by methods such as marking the region (R4) of the foreign substance (M) of the segmentation data on the image (G). Furthermore, if the model outputs incorrect segmentation data, the model can be precisely retrained to output correct segmentation data.

[0309] Meanwhile, the segmentation data is data in which the image (G) is partitioned into a plurality of overlapping regions (R), each comprising one or more regions (R) corresponding to one or more parts of the electrode portion (70) and one or more regions (R4) corresponding to one or more foreign substances (M) as described above, and when the aforementioned electrode data and foreign substance data each include at least a portion of the segmentation data, the position acquisition unit (400) can calculate the first position based on at least a portion of the first segmentation data included in the first electrode data, determine the presence of the aforementioned foreign substances (M) based on at least a portion of the first segmentation data included in the first foreign substance data, or calculate the third position, number, or area.

[0310] Accordingly, using a single segmentation data, it is possible to detect the edge (E) or corner (C) and foreign matter (M) of one or more parts of the electrode part (70). Accordingly, the first model can be machine-learned quickly and effectively at low cost, and the electrode vision inspection device (1) can be implemented simply and easily. In addition, training data can be generated simply and easily.

[0311] The position acquisition unit (400) can calculate the second position of one or more edges (E) or one or more corners (C) of each of one or more parts of the first electrode unit (70A) in the first image (G1) by applying a predefined rule or method (rule-based method).

[0312] Here, the predefined rule or method may be a method for detecting edges (E) or corners (C) that is predefined by the user. For example, it may be a method for detecting pixels as edges (E) or corners (C) in an image (G) where the brightness difference with adjacent pixels is greater than or equal to a reference value. The predefined rule or method may be a known method for detecting edges (E) or corners (C) that does not use machine learning.

[0313] [Evaluation Department]

[0314] The evaluation unit (500) can determine whether the first electrode unit (70A) is defective based on the first position. For example, the evaluation unit (500) can obtain or calculate the position (e.g., edge or corner position), length, width, or angle of one or more parts of the first electrode unit (70A) based on the first position, and compare the obtained or calculated value with a reference value to determine whether the first electrode unit (70A) is defective.

[0315] In this way, the electrode vision inspection device (1) may include a computation unit (10) and an image acquisition unit (60). The computation unit (10) may include a storage unit (100), a model learning unit (200), a model application unit (300), a position acquisition unit (400), and an evaluation unit (500). Accordingly, since the edge (E) or corner (C) of one or more parts of the first electrode part (70A) in the first image (G1) is detected using a machine-learned first model, the detection capability of the edge (E) or corner (C) is improved, and non-detection, over-detection, and false detection can be prevented. Accordingly, the accuracy of the defect determination of the electrode vision inspection device (1), which determines whether the electrode part (70) is defective based on the position of the edge (E) or corner (C) of one or more parts of the electrode part (70) detected in the image (G), can be improved. In particular, the accuracy of determining defects can be improved compared to when the electrode vision inspection device (1) does not use a machine-learned model, or when it uses a machine-learned model but uses a model trained to directly output a defect determination result from an input image (G).

[0316] In addition, the first model can be machine-learned accurately and precisely using training data that includes accurate electrode data. Accordingly, the accuracy of determining whether the electrode vision inspection device (1) is defective can be improved.

[0317] In addition, the first model can be easily and precisely verified by methods such as displaying the contents of the electrode data (e.g., edges (E) or corners (C)) on the image (G). Furthermore, if the first model outputs incorrect electrode data, the first model can be precisely retrained to output correct electrode data.

[0318] The evaluation unit (500) can obtain or calculate the position (e.g., corner or edge position), length, width, or angle of one or more parts of the first electrode unit (70A) based on the first position and compare it with a reference value to determine whether the first electrode unit (70A) is defective.

[0319] The evaluation unit (500) can calculate the length, width, slope, or protrusion length of the electrode tab (76) protruding outward from the electrode sheet (72) from a first position of the edge (E) or corner (C) of the electrode sheet (72, e.g., the unbleached portion) and / or the electrode tab (76). The evaluation unit (500) can determine whether the first electrode portion (70A) is defective by comparing the length, width, slope, or protrusion length, etc. with a reference value.

[0320] The evaluation unit (500) can calculate the width of the unused portion (A2), the length, width, slope, or the protruding length of the electrode tab (76) protruding outward from the unused portion (A2) from a first position of the edge (E) or corner (C) of the unused portion (A2) and / or the electrode tab (76). The evaluation unit (500) can determine whether the first electrode portion (70A) is defective by comparing the width, length, slope, or protruding length, etc. with a reference value.

[0321] The evaluation unit (500) can determine that the first electrode unit (70A) is defective if the electrode tab (76) is not present.

[0322] The evaluation unit (500) can further calculate the position (e.g., edge or corner position), width, length, or slope of the protective tape (78) from the first position of the edge (E) or corner (C) of the protective tape (78). The evaluation unit (500) can determine whether the first electrode unit (70A) is defective by comparing the above position, width, length, or slope with a reference value.

[0323] The evaluation unit (500) can determine that the first electrode unit (70A) is defective if the protective tape (78) is not present.

[0324] The evaluation unit (500) may determine that the first electrode unit (70A) is defective when one or more of the aforementioned foreign substances (M) are present, or determine whether the first electrode unit (70A) is defective based on the aforementioned third location, number, or area of ​​one or more of the foreign substances (M). For example, the evaluation unit (500) may determine whether the first electrode unit (70A) is defective by comparing the number or area of ​​one or more of the foreign substances (M) with a reference value. Here, the area may be the total area.

[0325] The evaluation unit (500) can produce a first result and a second result. Here, the first result is the result of the evaluation unit (500) determining whether the first electrode unit (70A) is defective based on the first position described above, and the second result is the result of the evaluation unit (500) determining whether the first electrode unit (70A) is defective based on the second position described above. The evaluation unit (500) can make a final determination of whether the first electrode unit (70A) is defective based on the first result and the second result.

[0326] Accordingly, the final determination of whether the first electrode part (70A) is defective is made based on two determination results calculated using two types of edge (E) or corner (C) detection methods (rule-based and machine learning), so the accuracy of determining whether the first electrode part (70A) is defective can be improved.

[0327] Here, the method of the final judgment may be selected or set by the user.

[0328] Accordingly, the final judgment method can be easily changed and the reliability of the final judgment can be easily adjusted.

[0329] For example, the method of final determination may be selected or set as a first method or a second method. The first method may be a method of finally determining the first electrode part (70A) as defective when both the first result and the second result are defective. The second method may be a method of finally determining the first electrode part (70A) as defective when at least one of the first result and the second result is defective.

[0330] Accordingly, the reliability of the final judgment can be easily adjusted.

[0331] In the case where the first result is defective and the second result is normal, the model learning unit (200) can machine learn the first model to output output data including electrode data corresponding to the second position (related to one or more edges (E) or one or more corners (C) of each of one or more parts of the first electrode unit (70A) when input data including the first image (G1) is input.

[0332] Accordingly, training data (output data) capable of training the first model can be acquired or produced using a predefined rule or method (rule-based method). Accordingly, the first model can be easily (re)trained. In addition, training data capable of training the first model can be automatically generated.

[0333] In this regard, the specifics are as follows.

[0334] If the first result is defective and the second result is normal, it can be seen that one or more edges (E) or corners (C) of the first electrode part (70A) in the first image (G1) are incorrectly detected by the machine learning method and correctly detected by a predefined rule or method (rule-based method). Accordingly, when the model learning part (200) inputs the first image (G1), it can (re)train the first model to output data (hereinafter, first data) related to the edges (E) or corners (C) detected by the rule-based method.

[0335] Here, the first data may be data (hereinafter referred to as the second data) for an edge (E) or corner (C) detected by a rule-based method, extracted or converted to correspond to the electrode data of the output data of the training data. For example, if the electrode data includes the aforementioned position data, the first data may be data corresponding to the position data extracted from the second data. Additionally, if the electrode data includes at least a portion of the aforementioned segmentation data, the first data may include at least a portion of data obtained by using the second data to divide the first image (G1) into multiple regions (R) that can be overlapped.

[0336] In the case where the first result is normal and the second result is defective, the evaluation unit (500) can provide the first image (G1) to the user.

[0337] Accordingly, the judgment results can be easily verified.

[0338] [Second Embodiment of Electrode Vision Inspection Device]

[0339] The electrode vision inspection device (1) according to the second embodiment may include a computation unit (10) and an image acquisition unit (60), similar to the electrode vision inspection device (1) according to the first embodiment described above. The computation unit (10) may include a storage unit (100), a model learning unit (200), a model application unit (300), a position acquisition unit (400), and an evaluation unit (500). The image acquisition unit (60) may acquire a first image (G1) that captures the first electrode unit (70A). Below, we will examine the differences from the electrode vision inspection device (1) according to the first embodiment.

[0340] The storage unit (100) may include one or more training data. Each training data may include input data and output data.

[0341] The input data may include an image (G) of the electrode part (70).

[0342] The output data may include foreign matter data. The foreign matter data may be data related to one or more foreign matter (M) that may be located on the surface of the electrode portion (70) in the image (G) (Figs. 6, 7).

[0343] The model learning unit (200) can machine learn the first model so that when input data of each training data is input, output data of each training data corresponding to the input data is output.

[0344] The model application unit (300) can input first input data including a first image (G1) into a machine-learned first model and output first output data including first foreign object data. The first foreign object data may correspond to foreign object data.

[0345] The location acquisition unit (400) can determine whether there is one or more foreign substances (M) on the surface of the first electrode part (70A) in the first image (G1) from the first foreign substance data, or calculate the third location, number, or area of ​​one or more foreign substances (M) on the surface of the first electrode part (70A).

[0346] The evaluation unit (500) may determine that the first electrode unit (70A) is defective when one or more foreign substances (M) are present, or determine whether the first electrode unit (70A) is defective based on the third position, number, or area of ​​one or more foreign substances (M).

[0347] Accordingly, since a machine-learned model is used to detect foreign matter (M) on the surface of the first electrode part (70A) in the first image (G1), the detection capability of the foreign matter (M) can be improved. Accordingly, the electrode vision inspection device (1) can accurately determine whether the electrode part (70) is defective based on the foreign matter (M) detected on the surface of the first electrode part (70A) in the first image (G1). In particular, the accuracy of determining whether the electrode vision inspection device (1) is defective can be improved compared to when the electrode vision inspection device (1) does not use a machine-learned model, or when a machine-learned model is used but a model trained to directly output a defect determination result from the input image (G) is used. In addition, even if a bubble (B) generated when attaching the protective tape (78) is present on the first electrode part (70A), the foreign matter (M) can be correctly detected by distinguishing it from the bubble (B) (Figs. 6, 7).

[0348] In addition, the model can be machine-learned accurately and precisely using training data that includes accurate foreign object data. Accordingly, the accuracy of determining whether the electrode vision inspection device (1) is defective can be improved.

[0349] In addition, the model can be easily and precisely verified by methods such as marking the foreign object (M) of the foreign object data on the image (G). Furthermore, if the model outputs incorrect foreign object data, the model can be precisely retrained to output correct foreign object data.

[0350] The aforementioned output data includes segmentation data, and the foreign material data may include at least a portion of the segmentation data. Here, the segmentation data may be data obtained by partitioning an image (G) into a plurality of overlapping regions (R), each comprising one or more regions (R4) corresponding to one or more foreign materials (M) that can be located on the surface of the electrode portion (70). Additionally, the aforementioned first output data includes first segmentation data corresponding to the segmentation data, and the first foreign material data may include at least a portion of the first segmentation data.

[0351] The location acquisition unit (400) can determine whether there is one or more foreign substances (M) on the surface of the first electrode unit (70A) based on at least a portion of the first segmentation data included in the first foreign substance data, or calculate the third location, number, or area.

[0352] Accordingly, the detection capability of the foreign substance (M) can be improved by using a machine-learned model to identify (divide) the region (R4) of the foreign substance (M) from the first image (G1). In particular, the foreign substance (M) can be accurately detected even if the location, shape, size, etc. of the foreign substance (M) change. In addition, even if a bubble (B) generated when attaching the protective tape (78) is present in the first electrode part (70A), the foreign substance (M) can be correctly detected by distinguishing it from the bubble (B) (Figs. 6, 7).

[0353] In addition, the model can be machine-learned accurately and precisely using training data that includes accurate segmentation data. Accordingly, the accuracy of determining whether the electrode vision inspection device (1) is defective can be improved.

[0354] In addition, since the foreign object region (R4) in the image (G) can be accurately and easily identified visually, correct training data can be easily generated.

[0355] In addition, the model can be easily and precisely verified by methods such as marking the region (R4) of the foreign substance (M) of the segmentation data on the image (G). Furthermore, if the model outputs incorrect segmentation data, the model can be precisely retrained to output correct segmentation data.

[0356] Meanwhile, matters not mentioned in relation to the electrode vision inspection device (1) according to the second embodiment can be inferred from the electrode vision inspection device (1) according to the first embodiment described above.

[0357] [First embodiment of electrode vision inspection method]

[0358] Referring to FIG. 8, the electrode vision inspection method (S900) according to the first embodiment may include a model learning process (S910), an image acquisition process (S920), a model application process (S930), a position acquisition process (S940), and an evaluation process (S950).

[0359] In the model learning process (S910), when the model learning unit (200) inputs input data of each training data, the first model can be machine-learned to output output data of each training data corresponding to the input data.

[0360] Here, the input data may include an image (G) of the electrode portion (70). The output data may include electrode data. The electrode data may be data related to one or more edges (E) or one or more corners (C) of each of one or more parts of the electrode portion (70) within the image (G).

[0361] In the image acquisition process (S920), the image acquisition unit (60) can acquire a first image (G1) of the first electrode unit (70A).

[0362] In the model application process (S930), the model application unit (300) can input first input data including a first image (G1) into the first model and output first electrode data corresponding to the electrode data.

[0363] In the position acquisition process (S940), the position acquisition unit (400) may acquire or calculate the first position of one or more edges (E) or one or more corners (C) of each of one or more parts of the first electrode part (70A) in the first image (G1) from the first electrode data (S942 of FIG. 9).

[0364] Here, when the electrode data includes the aforementioned position data and the first electrode data includes the aforementioned first position data, the position acquisition unit (400) can acquire the first position from the first position data.

[0365] Accordingly, since the first model is trained to directly output position data regarding the position of one or more edges (E) or one or more corners (C) of each of one or more parts of the first electrode part (70A), the accuracy of the position of each edge (E) or each corner (C) acquired by the position acquisition part (400) can be improved. Accordingly, the detection capability of each edge (E) or each corner (C) is improved, and non-detection, over-detection, and false detection can be prevented. In addition, since the first position data output from the first model can be easily set or converted to the first position, the electrode vision inspection method (S900) can be easily implemented at low cost.

[0366] In addition, the first model can be easily and precisely verified by methods such as displaying each edge (E) or each corner (C) of the location data on the image (G). Furthermore, if the first model outputs incorrect location data, the first model can be precisely retrained to output correct location data. Additionally, since the location of the correct edge (E) or corner (C) in the image (G) can be accurately and easily found visually, correct training data can be easily generated.

[0367] In addition, in the case where the output data includes the aforementioned segmentation data, the electrode data includes at least a portion of the segmentation data, the first output data includes first segmentation data corresponding to the segmentation data, and the first electrode data includes at least a portion of the first segmentation data, the position acquisition unit (400) can calculate the first position based on at least a portion of the first segmentation data included in the first electrode data. Here, the segmentation data may be data obtained by partitioning an image (G) into a plurality of overlapping regions (R) each including one or more regions (R) corresponding to one or more parts of the electrode unit (70).

[0368] Accordingly, after the corresponding region (R) for each part of the first electrode part (70A) is identified (divided) from the first image (G1), the edge (E) or corner (C) is detected from the identified (divided) region (R), so the detection capability of the edge (E) or corner (C) is improved and non-detection, over-detection, and false detection can be prevented. In particular, the edge (E) or corner (C) can be accurately detected even if the position or shape of one or more parts of the electrode part (70) changes.

[0369] In addition, the first model can be machine-learned accurately and precisely using training data that includes accurate segmentation data. Accordingly, the accuracy of determining defects in the electrode vision inspection method (S900) can be improved.

[0370] In addition, since the region (R) of the electrode part (70) in the image (G) can be accurately and easily identified visually, correct training data can be easily generated.

[0371] In addition, the first model can be easily and precisely verified by methods such as displaying each region (R) of the segmentation data on the image (G). Furthermore, if the first model outputs incorrect segmentation data, the first model can be precisely retrained to output correct segmentation data.

[0372] In addition, the position acquisition unit (400) can calculate the second position of one or more edges (E) or one or more corners (C) of each of one or more parts of the first electrode unit (70A) in the first image (G1) by applying a predefined rule or method (S944 of FIG. 9).

[0373] In addition, in the case where the output data of the training data includes foreign object data related to one or more foreign objects (M) within the image (G) and the first output data includes first foreign object data corresponding to the foreign object data, the position acquisition unit (400) can determine whether one or more foreign objects (M) exist from the first foreign object data or calculate a third position, number, or area of ​​one or more foreign objects (M).

[0374] Accordingly, since a machine-learned model is used to detect foreign matter (M) on the surface of the electrode part (70) in the image (G), the detection capability of the foreign matter (M) can be improved. Accordingly, the electrode vision inspection method (S900) can accurately determine whether the electrode part (70) is defective based on the foreign matter (M) detected on the surface of the electrode part (70) in the image (G). In particular, the accuracy of determining whether the electrode part is defective can be improved compared to when the electrode vision inspection method (S900) does not use a machine-learned model, or when a machine-learned model is used but a model trained to directly output a result of determining whether the defect is defective from the input image (G) is used. In addition, even if bubbles (B) generated when attaching the protective tape (78) are present on the electrode part (70), the foreign matter (M) can be correctly detected by distinguishing it from the bubbles (B) (Figs. 6, 7).

[0375] In addition, the model can be machine-learned accurately and precisely using training data that includes accurate foreign object data. Accordingly, the accuracy of determining defects in the electrode vision inspection method (S900) can be improved.

[0376] In addition, the model can be easily and precisely verified by methods such as displaying foreign object data on the image (G). Furthermore, if the model outputs incorrect foreign object data, the model can be precisely retrained to output correct foreign object data.

[0377] In the evaluation process (S950), the evaluation unit (500) can determine whether the first electrode unit (70A) is defective based on the first position.

[0378] In this way, the electrode vision inspection method (S900) may include a model learning process (S910), an image acquisition process (S920), a model application process (S930), a position acquisition process (S940), and an evaluation process (S950). Accordingly, since the edge (E) or corner (C) of one or more parts of the first electrode part (70A) in the first image (G1) is detected using a machine-learned first model, the detection capability of the edge (E) or corner (C) is improved, and non-detection, over-detection, and false detection can be prevented. Accordingly, the accuracy of determining whether the electrode part (70) is defective can be improved in the electrode vision inspection method (S900), which determines whether the electrode part (70) is defective based on the position of the edge (E) or corner (C) of one or more parts of the electrode part (70) detected in the image (G). In particular, the accuracy of determining defects can be improved compared to when the electrode vision inspection method (S900) does not use a machine-learned model, or when a machine-learned model is used but a model is trained to directly output a defect determination result from an input image (G).

[0379] In addition, the first model can be machine-learned accurately and precisely using training data that includes accurate electrode data. Accordingly, the accuracy of determining defects in the electrode vision inspection method (S900) can be improved.

[0380] In addition, the first model can be easily and precisely verified by methods such as displaying the contents of the electrode data (e.g., edges (E) or corners (C)) on the image (G). Furthermore, if the first model outputs incorrect electrode data, the first model can be precisely retrained to output correct electrode data.

[0381] Here, the evaluation unit (500) can produce a first result determining whether the first electrode unit (70A) is defective based on the first position and a second result determining whether the first electrode unit (70A) is defective based on the second position (S952, S954 of FIG. 9). The evaluation unit (500) can finally determine whether the first electrode unit (70A) is defective based on the first result and the second result (S956 of FIG. 9).

[0382] Accordingly, the final determination of whether the first electrode part (70A) is defective is made based on two determination results calculated using two types of edge (E) or corner (C) detection methods (rule-based and machine learning), so the accuracy of determining whether the first electrode part (70A) is defective can be improved.

[0383] In addition, the method of final determination here may be selected or set by the user (S956 of FIG. 9). For example, the method of final determination may be selected or set as a first method (AND operation) or a second method (OR operation). The first method may be a method of finally determining the first electrode part (70A) as defective when both the first result and the second result are defective. The second method may be a method of finally determining the first electrode part (70A) as defective when at least one of the first result and the second result is defective.

[0384] In addition, in the case where the first result is defective and the second result is normal, the model learning unit (200) can machine learn the first model to output output data including electrode data corresponding to the second position (related to one or more edges (E) or one or more corners (C) of each of one or more parts of the first electrode unit (70A) when input data including the first image (G1) is input.

[0385] On the other hand, if the first result is normal and the second result is defective, the evaluation unit (500) can provide the first image (G1) to the user.

[0386] In addition, here, the evaluation unit (500) may determine that the first electrode unit (70A) is defective when one or more foreign substances (M) are present, or determine whether the first electrode unit (70A) is defective based on the third location, number, or area of ​​one or more foreign substances (M).

[0387] Meanwhile, matters not mentioned in relation to the electrode vision inspection method (S900) according to the first embodiment can be inferred from the aforementioned electrode vision inspection device (1).

[0388] [Second Embodiment of Electrode Vision Inspection Method]

[0389] The electrode vision inspection method (S900) according to the second embodiment may include a model learning process (S910), an image acquisition process (S920), a model application process (S930), a position acquisition process (S940), and an evaluation process (S950), similar to the electrode vision inspection method (S900) according to the first embodiment described above. We will examine the differences from the electrode vision inspection method (S900) according to the first embodiment.

[0390] In the model learning process (S910), when the model learning unit (200) inputs input data of each training data, the first model can be machine-learned to output output data of each training data corresponding to the input data.

[0391] Here, the input data may include an image (G) of the electrode portion (70). The output data may include foreign matter data. The foreign matter data may be data related to one or more foreign matter (M) that may be located on the surface of the electrode portion (70) within the image (G).

[0392] In the image acquisition process (S920), the image acquisition unit (60) can acquire a first image (G1) of the first electrode unit (70A).

[0393] In the model application process (S930), the model application unit (300) can input first input data including a first image (G1) into the first model and output first output data including first foreign material data corresponding to foreign material data.

[0394] In the position acquisition process (S940), the position acquisition unit (400) can determine whether there is one or more foreign substances (M) on the surface of the first electrode part (70A) within the first image (G1) from the first foreign substance data, or calculate the third position, number, or area of ​​one or more foreign substances (M) on the surface of the first electrode part (70A).

[0395] In the evaluation process (S950), if one or more foreign substances (M) are present, the evaluation unit (500) may determine that the first electrode unit (70A) is defective, or determine whether the first electrode unit (70A) is defective based on the third location, number, or area of ​​one or more foreign substances (M).

[0396] Accordingly, since a machine-learned model is used to detect foreign matter (M) on the surface of the first electrode part (70A) in the first image (G1), the detection capability of the foreign matter (M) can be improved. Accordingly, the electrode vision inspection method (S900) can accurately determine whether the electrode part (70) is defective based on the foreign matter (M) detected on the surface of the first electrode part (70A) in the first image (G1). In particular, the accuracy of determining defects can be improved compared to when the electrode vision inspection method (S900) does not use a machine-learned model, or when a machine-learned model is used but a model trained to directly output a defect determination result from the input image (G) is used. In addition, even if bubbles (B) generated when attaching the protective tape (78) are present on the first electrode part (70A), the foreign matter (M) can be correctly detected by distinguishing it from the bubbles (B) (Figs. 6, 7).

[0397] In addition, the model can be machine-learned accurately and precisely using training data that includes accurate foreign object data. Accordingly, the accuracy of determining defects in the electrode vision inspection method (S900) can be improved.

[0398] In addition, the model can be easily and precisely verified by methods such as marking the foreign object (M) of the foreign object data on the image (G). Furthermore, if the model outputs incorrect foreign object data, the model can be precisely retrained to output correct foreign object data.

[0399] Meanwhile, matters not mentioned in relation to the electrode vision inspection method (S900) according to the second embodiment can be inferred from the electrode vision inspection method (S900) according to the first embodiment described above or the electrode vision inspection device (1) described above.

[0400] The embodiments described above should be understood as exemplary in all respects and not limiting, and the scope of the invention will be defined by the claims set forth below rather than by the detailed description above. Furthermore, the meaning and scope of the claims set forth below, as well as all modifications and variations derived from equivalents thereof, should be interpreted as being included within the scope of the invention.

[0401] Although the present invention has been described above with reference to the illustrated drawings, the present invention is not limited by the embodiments and drawings disclosed in this specification, and it is obvious that various modifications can be made by a person skilled in the art within the scope of the technical concept of the present invention. Furthermore, even if the effects of the configuration according to the present invention were not explicitly described while describing the embodiments of the present invention above, it is natural to acknowledge that the effects predictable by said configuration should also be recognized.

Claims

1. A storage unit (100) storing one or more training data, each comprising input data including an image (G) of an electrode portion (70) and output data including electrode data associated with one or more edges (E) or one or more corners (C) of each of one or more parts of the electrode portion (70) within the image (G); A model learning unit (200) that machine learns a first model to output the output data corresponding to the input data when the input data of each of the above training data is input; An image acquisition unit (60) that acquires a first image (G1) of the first electrode part (70A); A model application unit (300) that inputs first input data including the first image (G1) into the first model and outputs first output data including first electrode data corresponding to the electrode data; A position acquisition unit (400) for acquiring or calculating a first position of one or more edges (E) or one or more corners (C) of each of one or more parts of the first electrode portion (70A) within the first image (G1) from the first electrode data; and The evaluation unit (500) for determining whether the first electrode part (70A) is defective based on the first position above, Electrode vision inspection device.

2. In Claim 1, The above-mentioned first model is an electrode vision inspection device, which is a deep learning model.

3. In claim 1 or claim 2, The above electrode data includes position data regarding the location of one or more edges (E) or one or more corners (C) of each of the one or more parts of the electrode portion (70) within the image (G), and The first electrode data includes first position data regarding the position of one or more edges (E) or one or more corners (C) of each of the one or more parts of the first electrode part (70A) within the first image (G1), and The above position acquisition unit (400) is an electrode vision inspection device that acquires the first position from the first position data.

4. In claim 1 or claim 2, The above output data includes segmentation data in which the image (G) is partitioned into a plurality of overlapping regions (R), each comprising one or more regions (R) corresponding to one or more parts of the electrode portion (70). The electrode data above includes at least a portion of the segmentation data, and The first output data above includes first segmentation data corresponding to the segmentation data, and The first electrode data includes at least a portion of the first segmentation data, and The above position acquisition unit (400) is an electrode vision inspection device that calculates the first position based on at least a portion of the first segmentation data included in the first electrode data.

5. In Claim 4, Each of the above electrode portion (70) and the first electrode portion (70A) includes an electrode sheet (72) and an electrode tab (76) that can be coupled to the electrode sheet (72). Each of the above electrode portion (70) and the first electrode portion (70A) comprises one or more portions including the electrode sheet (72) and the electrode tab (76). The above position acquisition unit (400) calculates the first position of one or more edges (E) or one or more corners (C) of each of at least one part of the electrode sheet (72) and the electrode tab (76) based on at least a portion of the first segmentation data included in the first electrode data, an electrode vision inspection device.

6. In Claim 5, The electrode sheet (72) comprises a retaining portion (A1) on which an electrode active material (74) is applied to an electrode foil (73) and a non-retaining portion (A2) on which the electrode active material (74) is not applied to the electrode foil (73). The above electrode tab (76) can be coupled to the above unpaired portion (A2), and Each of the above electrode portion (70) and the first electrode portion (70A) includes one or more portions, and the above electrode portion (A2) and the electrode tab (76). The above position acquisition unit (400) calculates the first position of one or more edges (E) or one or more corners (C) of each of at least one part of the above-mentioned portion (A2) and the electrode tab (76) based on at least a portion of the above-mentioned first segmentation data included in the above-mentioned first electrode data, electrode vision inspection device.

7. In claim 5 or claim 6, Each of the above electrode portion (70) and the first electrode portion (70A) further includes a protective tape (78) that can be attached to the electrode sheet (72) or electrode tab (76), and Each of the above electrode portion (70) and the first electrode portion (70A) further includes the above protective tape (78), and The above position acquisition unit (400) calculates the first position of one or more edges (E) or one or more corners (C) of each of at least one part of the electrode sheet (72), electrode tab (76) and protective tape (78) based on at least a portion of the first segmentation data included in the first electrode data, an electrode vision inspection device.

8. In any one of claims 1 to 7, The above position acquisition unit (400) calculates a second position of one or more edges (E) or one or more corners (C) of each of the one or more parts of the first electrode part (70A) in the first image (G1) by applying a predefined rule or method, and The above evaluation unit (500) calculates a first result determining whether the first electrode part (70A) is defective based on the first position and a second result determining whether the first electrode part (70A) is defective based on the second position, and finally determines whether the first electrode part (70A) is defective based on the first result and the second result, an electrode vision inspection device.

9. In Claim 8, The above method of final determination can be selected or set as the first method or the second method, and The above first method is a method for finally determining the above first electrode part (70A) as defective when both the above first result and the above second result are defective, and The above second method is an electrode vision inspection device that determines the first electrode part (70A) as defective when at least one of the first result and the second result is defective.

10. In claim 8 or claim 9, When the first result is defective and the second result is normal, the model learning unit (200) machine learns the first model to output the output data including the electrode data associated with each of the one or more edges (E) or one or more corners (C) of the first electrode unit (70A) corresponding to the second position when the input data including the first image (G1) is input.

11. In any one of claims 1 to 10, One or more foreign substances (M) may be located on the surface of the electrode portion (70) and the first electrode portion (70A), and The output data of each of the above training data includes foreign material data related to one or more foreign materials (M) within the image (G), and The above first output data includes first foreign material data corresponding to the above foreign material data, and The above location acquisition unit (400) determines whether there is one or more foreign substances (M) from the first foreign substance data, or calculates a third location, number, or area of ​​the one or more foreign substances (M). The above evaluation unit (500) determines the first electrode unit (70A) as defective when one or more foreign substances (M) are present, or determines whether the first electrode unit (70A) is defective based on the third position, number, or area of ​​the one or more foreign substances (M), an electrode vision inspection device.

12. In Claim 11, The above output data includes segmentation data in which the image (G) is partitioned into a plurality of overlapping regions (R) each comprising one or more regions (R4) corresponding to one or more foreign substances (M). The above foreign material data includes at least a portion of the above segmentation data, and The first output data above includes first segmentation data corresponding to the segmentation data, and The first foreign material data above includes at least a portion of the first segmentation data, and The above position acquisition unit (400) determines the presence of one or more foreign substances (M) based on at least a portion of the first segmentation data included in the first foreign substance data, or calculates the third position, number, or area, electrode vision inspection device.

13. In Claim 12, The above segmentation data is data obtained by partitioning the image (G) into a plurality of overlapping regions (R), each comprising one or more regions (R) corresponding to one or more parts of the electrode portion (70) and one or more regions (R4) corresponding to one or more foreign substances (M). The electrode data above includes at least a portion of the segmentation data, and The above position acquisition unit (400) is an electrode vision inspection device that calculates the first position based on at least a portion of the first segmentation data included in the first electrode data.

14. A storage unit (100) storing one or more training data, each comprising input data including an image (G) of an electrode portion (70) and output data including foreign matter data related to one or more foreign matter (M) that may be located on the surface of the electrode portion (70) within the image (G); A model learning unit (200) that machine learns a first model to output the output data corresponding to the input data when the input data of the above training data is input; An image acquisition unit (60) that acquires a first image (G1) of the first electrode part (70A); A model application unit (300) that inputs first input data including the first image (G1) into the first model and outputs first output data including first foreign matter data corresponding to the foreign matter data; A position acquisition unit (400) that determines whether there is one or more foreign substances (M) on the surface of the first electrode part (70A) within the first image (G1) from the first foreign substance data, or calculates a third position, number, or area of ​​one or more foreign substances (M) on the surface of the first electrode part (70A); and The evaluation unit (500) includes a first electrode part (70A) that determines the first electrode part (70A) as defective when one or more foreign substances (M) are present, or determines whether the first electrode part (70A) is defective based on the third position, number, or area of ​​one or more foreign substances (M). Electrode vision inspection device.

15. In Claim 14, The above output data includes segmentation data in which the image (G) is partitioned into a plurality of overlapping regions (R), each comprising one or more regions (R4) corresponding to one or more foreign substances (M) that can be located on the surface of the electrode portion (70). The above foreign material data includes at least a portion of the above segmentation data, and The first output data above includes first segmentation data corresponding to the segmentation data, and The first foreign material data above includes at least a portion of the first segmentation data, and The above position acquisition unit (400) determines whether there is one or more foreign substances (M) on the surface of the first electrode unit (70A) based on at least a portion of the first segmentation data included in the first foreign substance data, or calculates the third position, number, or area, electrode vision inspection device.

16. An electrode vision inspection method (S900) using the electrode vision inspection device (1) of claim 1 or claim 2, A model learning process (S910) in which the above model learning unit (200) inputs the input data of each of the training data, which includes the image (G) of the electrode unit (70), and outputs the output data of each of the training data, which includes the electrode data corresponding to the input data and related to one or more edges (E) or one or more corners (C) of each of one or more parts of the electrode unit (70) within the image (G); An image acquisition process (S920) in which the image acquisition unit (60) acquires the first image (G1) that captures the first electrode unit (70A); A model application process (S930) in which the above model application unit (300) inputs the first input data including the first image (G1) to the first model and outputs the first electrode data corresponding to the electrode data; A position acquisition process (S940) in which the position acquisition unit (400) acquires or calculates the first position of one or more edges (E) or one or more corners (C) of each of the one or more parts of the first electrode part (70A) within the first image (G1) from the first electrode data; and The evaluation unit (500) includes an evaluation process (S950) for determining whether the first electrode unit (70A) is defective based on the first position. Electrode vision inspection method.

17. In Claim 16, The above electrode data includes position data regarding the location of one or more edges (E) or one or more corners (C) of each of the one or more parts of the electrode portion (70) within the image (G), and The first electrode data includes first position data regarding the position of one or more edges (E) or one or more corners (C) of each of the one or more parts of the first electrode part (70A) within the first image (G1), and An electrode vision inspection method in which, in the above position acquisition process (S940), the position acquisition unit (400) acquires the first position from the first position data.

18. In Claim 16, The above output data includes segmentation data in which the image (G) is partitioned into a plurality of overlapping regions (R), each comprising one or more regions (R) corresponding to one or more parts of the electrode portion (70). The electrode data above includes at least a portion of the segmentation data, and The first output data above includes first segmentation data corresponding to the segmentation data, and The first electrode data includes at least a portion of the first segmentation data, and An electrode vision inspection method in which, in the above position acquisition process (S940), the position acquisition unit (400) calculates the first position based on at least a portion of the first segmentation data included in the first electrode data.

19. In any one of claims 16 to 18, In the above position acquisition process (S940), the position acquisition unit (400) calculates a second position of an edge (E) or corner (C) of one or more parts of the first electrode part (70A) in the first image (G1) by applying a predefined rule or method, and In the above evaluation process (S950), the evaluation unit (500) calculates a first result determining whether the first electrode unit (70A) is defective based on the first position and a second result determining whether the first electrode unit (70A) is defective based on the second position, and finally determines whether the first electrode unit (70A) is defective based on the first result and the second result, an electrode vision inspection device.

20. In any one of claims 16 to 19, One or more foreign substances (M) may be located on the surface of the electrode portion (70) and the first electrode portion (70A), and The output data of each of the above training data includes foreign material data related to one or more foreign materials (M) within the image (G), and The above first output data includes first foreign material data corresponding to the above foreign material data, and In the above location acquisition process (S940), the location acquisition unit (400) determines whether there is one or more foreign substances (M) from the first foreign substance data, or calculates a third location, number, or area of ​​the one or more foreign substances (M). An electrode vision inspection method in which, in the above evaluation process (S950), the evaluation unit (500) determines the first electrode unit (70A) as defective when one or more foreign substances (M) are present, or determines whether the first electrode unit (70A) is defective based on the third position, number, or area of ​​the one or more foreign substances (M).

Images