JP7308301B2 - Inspection device, bag making system, judgment model, and inspection method - Google Patents

Description

The present disclosure relates to an inspection device that inspects bags, a judgment model used for inspection, a bag-making system including the inspection device, and an inspection method.

For the purpose of packaging various products and packing liquid products for transportation, these products are put into bags. Such bags are inspected for defects such as flaws and contamination before the product is put into the bag. Conventionally, in such inspections, except for some simple inspections, visual inspections by inspection workers are common. On the other hand, for example, Patent Literature 1 discloses a package inspection apparatus that detects defects in a sealed portion of a package, which is a type of bag. The inspection apparatus disclosed in Patent Document 1 compares an inspection image with a reference image, and detects defects in the sealing portion by pattern matching or the like.

JP 2017-083403 A

However, the inspection apparatus disclosed in Patent Literature 1 may not be able to perform accurate inspection.

Therefore, the present disclosure provides an inspection apparatus and the like that can perform inspections more accurately.

In order to solve the above problems, an inspection apparatus according to an aspect of the present disclosure is a pair of sheets in which a first sheet having a first pattern and a second sheet having at least a part of it being translucent are stacked. an inspection device for inspecting a pair of sheets forming a bag body by joining respective ends of the first sheet and the second sheet, wherein the pair of sheets is imaged from the side of the second sheet. at least one of a non-defective sheet pair determined to be good out of good or defective in the inspection, and a defective sheet pair determined to be defective out of good or defective in the inspection a judgment unit that judges at least one of good and bad of the pair of sheets corresponding to the acquired image by a judgment model constructed by machine learning using a teacher image captured from the second sheet side; , provided.

In addition, a bag-making system according to an aspect of the present disclosure is a pair of sheets in which a first sheet having a first pattern and a second sheet having at least a portion thereof are translucent are stacked, wherein the first sheet and A forming device for forming a bag by joining respective ends of the second sheet, and the inspection device described above.

Further, a determination model according to an aspect of the present disclosure is a determination model that is included in a program executed by a computer and that outputs output data determined based on an image that is input as input data, the first pattern and a second sheet at least partially translucent, wherein the ends of the first sheet and the second sheet are joined to each other Determining at least one of good and bad of the pair of sheets corresponding to the acquired image when an image of the pair of sheets forming the bag body captured from the side of the second sheet is input as the input data. and output the determination result as the output data.

Further, an inspection method according to an aspect of the present disclosure is a pair of sheets in which a first sheet having a first pattern and a second sheet having at least a part thereof having translucency are stacked, wherein the first sheet and an inspection method for inspecting a pair of sheets formed by joining the ends of the second sheets to form a bag, wherein an acquisition step of acquiring an image of the pair of sheets taken from the side of the second sheet and at least one of a pair of non-defective sheets determined to be good out of good or defective in the inspection and a pair of defective sheets determined to be defective out of good or defective from the second sheet side. a determination step of determining at least one of good and bad of the pair of sheets corresponding to the acquired image by a determination model constructed by machine learning using the captured teacher image.

According to the present disclosure, an inspection device and the like that can perform inspections more accurately are provided.

FIG. 1 is a schematic diagram showing the appearance of an inspection apparatus according to an embodiment. FIG. 2 is a diagram for explaining the sheet pair according to the embodiment. FIG. 3A is a first diagram illustrating a superimposed pattern according to the embodiment; FIG. FIG. 3B is a second diagram illustrating superimposed patterns according to the embodiment. FIG. 3C is a third diagram illustrating superimposed patterns according to the embodiment. FIG. 4 is a block diagram showing the inspection device according to the embodiment. FIG. 5 is a diagram for explaining an imaging device according to an embodiment. FIG. 6 is a diagram explaining the content of an image acquired in the embodiment. FIG. 7 is a diagram for explaining a determination model used in the determination unit according to the embodiment; FIG. 8 is a flow chart showing the operation of the bag making system according to the embodiment. FIG. 9 is a block diagram showing a generation device and an inspection device according to the embodiment. FIG. 10 is a flow chart showing the operation of the generator. FIG. 11 is an overview diagram showing the appearance of a bag-making system according to a modification of the embodiment. FIG. 12 is a flow chart showing the operation of the bag making system according to the modification of the embodiment.

(Knowledge leading to the present disclosure)
The technique of forming a bag by stacking two sheets and joining the ends together is used in the packaging of many products. For example, a four-sided seal, which is obtained by stacking two rectangular sheets of approximately the same size and joining them at the four sides, is generally widely used because it is a package suitable for transportation with little waste. The bag formed by sealing (that is, joining) the ends in this way is supplied to the product maker and shipped with the product enclosed therein. That is, the bag body has an opening for containing the product to be enclosed, and the product is packaged by further sealing the opening after the product is contained.

Conventionally, the package containing the product in this manner was further loaded into an exterior box or the like. However, in recent years, the durability of the sheet material used for such packages has been improved, and due to the consideration of the environmental burden of using an outer box etc., the product is shipped in a state of being only enclosed in the package. is also becoming popular.

Here, in order for the product to be shipped only by being enclosed in the package, it is necessary to form various kinds of information as patterns on the sheet used for the package. Specifically, on the sheet used for the package, the product name and manufacturer name of the enclosed product are formed using letters and numbers, etc., and the manufacturer's logo and decorative patterns are formed using figures, symbols, patterns, etc. is formed.

In addition, defects such as scratches, fisheyes, and pinholes on the sheet used for the package (the entire surface including the part where the edge is sealed), and foreign matter entering the sheet may be detected in the shipped product. Inspections for the presence or absence of these abnormalities are carried out so as not to occur. The inspection may be (i) after the product is packaged, (ii) before the product is packaged, or (iii) both. As described above, when the packaging bag and the product are manufactured by different manufacturers or manufactured on different production lines, inspection is performed at least before the product is enclosed. ((ii) or (iii) above) is the mainstream.

In such inspections, there are various variations of abnormalities based on various causes of abnormalities, and automation is considered to be difficult. For this reason, until recently, inspections have been conducted manually by inspectors and the like.

In recent years, some of such inspections have been replaced by automatic inspections using images, etc., but the accuracy of such inspections may not be sufficient. Therefore, the present disclosure is made in view of the above, and aims to provide an inspection apparatus and the like that can perform inspections more accurately.

Embodiments of the present disclosure will be described below with reference to the drawings. It should be noted that all of the embodiments described below represent comprehensive or specific examples of the present disclosure. Therefore, numerical values, components, arrangement positions and connection forms of components, steps, order of steps, and the like shown in the following embodiments are examples and are not intended to limit the present disclosure. Therefore, among constituent elements in the following embodiments, constituent elements not described in independent claims of the present disclosure will be described as optional constituent elements.

Each figure is a schematic diagram and is not necessarily strictly illustrated. Therefore, scales and the like are not always the same in each drawing. In each figure, the same reference numerals are assigned to substantially the same configurations, and duplicate descriptions are omitted or simplified.

(Embodiment)
[composition]
First, the configuration of an inspection apparatus according to an embodiment and the configuration of a bag to be inspected by the inspection apparatus will be described with reference to FIGS. 1 to 6. FIG. FIG. 1 is a schematic diagram showing the appearance of an inspection apparatus according to an embodiment. In FIG. 1, in addition to the inspection device 100, an imaging device 200 and a transport path 400 are illustrated. FIG. 1 also shows a first light source 207 and a second light source 208 used for imaging the sheet pair 13 (see FIG. 2 described later) by the imaging device 200 .

As shown in FIG. 1, the inspection apparatus 100 according to the present embodiment uses an image 209 (see FIG. 6, which will be described later) captured by the imaging device 200 to detect the bag body 14 transported on the transport surface of the transport path 400. It is a device that determines at least one of the quality and the quality of the sheet pair 13 in. A plurality of bags 14 are formed in advance by a forming device and arranged on a conveying path 400 at substantially regular intervals, and are positioned below a first camera 201, a second camera 202, and a third camera 203 provided in an imaging device 200. Inspection by the inspection apparatus 100 is performed in order from the conveyed items. That is, in this embodiment, the sheet pair 13 after the bag body 14 has been formed is automatically and continuously inspected.

Here, the “determination of at least one of good or bad” of the sheet pair 13 in the present disclosure means “non-defective product” (that is, the sheet pair 13 that is determined to be good) and “defective product” (that is, the sheet that is determined to be defective). It is not intended only for "pass/fail judgment" in which the judgment of both pair 13) is directly performed and sorted. For example, "determining at least one of good or bad" includes the selection of "good or bad" that directly performs the operation of determining "good" and does not directly determine "defective". This is the concept of In addition, "determining at least one of good or bad" includes the selection of "defective or not" that directly performs the operation of determining "defective product" and does not directly determine "defective product". This is the concept of Which of these determinations is applied is appropriately selected and used according to conditions such as required accuracy, yield, and cost.

The conveying path 400 is, for example, a belt conveyor or other device that conveys the placed object (that is, the bag body 14) by moving it in a predetermined direction by rotating a power source or the like. The conveying path 400 may be realized by using a suction conveyor since the bag body 14 to be conveyed is in the form of a sheet. By realizing the conveying path 400 with a suction conveyor, the bag body 14 can be sucked and held on the conveying surface. At this time, since the undulation of the bag body 14 and the like is suppressed, a better image 209 is obtained when the imaging device 200 captures the image.

However, the transport path 400 is composed of at least two transport paths 400 spaced apart at locations corresponding to the first light sources 207 so that the light emitted from the first light sources 207 reaches the bag body 14 . For this reason, a configuration different from that of the suction conveyor (for example, a belt conveyor on which the object to be transported is simply placed, etc.) is employed at the separated portion of the transport path 400 . Light emitted from the first light source 207 is transmitted while the bag body 14 is conveyed across the spaced location, and is received by the imaging device 200 to capture an image 209 .

Now, the bag body 14 will be described. FIG. 2 is a diagram for explaining the sheet pair according to the embodiment. 2(a) shows the sheet pair 13 before the bag 14 is formed, and FIG. 2(b) shows the sheet pair 13 after the bag 14 is formed. In FIG. 2B, portions corresponding to the joined ends 11a and 12a are indicated by dot hatching.

As described above, the bag body 14 in the present embodiment includes the film-like first sheet 11 and the film-like second sheet 12 that form the sheet pair 13, and the end portion 11a of the first sheet 11 and the second sheet 12 are overlapped. It is formed by joining the ends 12 a of the sheet 12 . In this embodiment, an example will be described in which the bag body 14 is formed by stacking the first sheet 11 and the second sheet 12 of the same size and having the same size and joining the ends 11a and 12a. It should be noted that the present disclosure is not limited to only the sheet pair 13 forming the bag 14 having such a shape, but is applicable to forming the bag 14 using the first sheet 11 and the second sheet 12 having any shape. applicable to

In addition, in the present embodiment, as described above, the bag body 14 having an opening for inserting a separate product is formed. It is assumed that the bag body 14 in the state of is inspected.

The same material may be used for the first sheet 11 and the second sheet 12, or different materials may be used. For example, the first sheet 11 and the second sheet 12 are both resin sheet materials made of nylon, polyethylene, polypropylene, or the like. Also, for example, the first sheet 11 may be a resin sheet material composed of nylon, polyethylene, polypropylene, or the like, and the second sheet 12 may be a non-woven fabric in which various natural or synthetic fibers are accumulated.

At least one of the first sheet 11 and the second sheet 12 (both in the figure) is formed with a perforation 15 for opening the formed bag 14 . By splitting the bag 14 along such perforations 15, it is possible to control the shape of the opening when the bag 14 is unsealed, and to prevent accidents such as dropping the enclosed product.

Here, the first pattern 11b and the second pattern 12b formed on the first sheet 11 and the second sheet 12 will be described. FIG. 3A is a first diagram illustrating a superimposed pattern according to the embodiment; FIG. FIG. 3B is a second diagram illustrating superimposed patterns according to the embodiment. FIG. 3C is a third diagram illustrating superimposed patterns according to the embodiment. FIGS. 3A to 3C show left-right reversed printing of "agent A" as an example of the first pattern 11b, and show printing of "antiviral drug" as an example of the second pattern 12b. In the figure, the inside of the "Agent A" and "Antiviral drug" printed is a portion that exhibits a semi-light-shielding property with a lower transmittance than the outside. It is shown in white for convenience. For example, the above-described print is formed as colored characters on the transparent first sheet 11 and the second sheet 12 by a method such as printing. For example, it is printed with a color such as red or blue, which has a high contrast ratio with respect to the background and shadow of the transparent sheet.

As shown in FIGS. 3A to 3C, a first pattern 11b and a second pattern 12b are formed in advance on the first sheet 11 and the second sheet 12, respectively. Here, at least a portion of the first sheet 11 has translucency. Having translucency means, for example, being transparent, and means a state in which the opposite side can be visually recognized through the first sheet 11 .

As indicated by hatching in FIG. 3A, the first pattern 11b and the second pattern 12b overlap to form a superimposed pattern 14a with relatively low transmittance.

Here, as shown in FIG. 3B, the manner in which the first pattern 11b and the second pattern 12b are overlapped differs depending on the arrangement positions of the first sheet 11 and the second sheet 12 . Specifically, a pseudo-abnormal pattern 14b, which is a narrow superimposed pattern 14a, is formed in part of the superimposed pattern 14a. Such a pseudo abnormal pattern 14b is formed simply by overlapping the first pattern 11b and the second pattern 12b, but the flaws, fisheyes, and pinholes formed in the sheet pair 13 and the sheet pair 13 It is difficult to distinguish between mixed foreign matter and the like by threshold judgment based on size.

In particular, as described above, depending on the arrangement positions of the first pattern 11b and the second pattern 12b, the formation position of the pseudo abnormal pattern 14b greatly changes. Specifically, as shown in FIGS. 3B and 3C, the placement positions of the first pattern 11b and the second pattern 12b are only slightly different, and the pseudo-abnormal pattern 14b extends from the right side of the character "medicine" to the left side. , the formation position changes. In such a case, as shown in FIG. 3C, when abnormal marks 16 such as scratches, fisheyes, pinholes, and foreign substances are actually formed, it is difficult to distinguish these from the pseudo abnormal patterns 14b. . In particular, when the abnormal mark 16 is formed so as to overlap the first pattern 11b (see FIG. 3C), and when the perforation 15 overlaps the projected sewing pattern, the distinction becomes even more difficult.

FIG. 4 is a block diagram showing the inspection device according to the embodiment. As described with reference to FIG. 1, the inspection apparatus 100 in the present embodiment is connected to the imaging device 200, acquires the image 209 from the imaging device 200, and determines the quality of the sheet pair 13 using the image 209. and at least one of failure is determined.

As shown in FIG. 4 , the inspection apparatus 100 includes an acquisition section 102 and a determination section 101 . The acquisition unit 102 is a processing unit that acquires the image 209 from the imaging device 200 . The acquisition unit 102 performs processing such as converting the acquired image 209 into a format that can be processed by the determination unit 101 described later as necessary.

The determination unit 101 is a processing unit that uses the image 209 acquired by the acquisition unit 102 to determine at least one of whether the sheet pair 13 corresponding to the image 209 is good or bad. That is, based on the acquired image 209, the determination unit 101 determines whether the abnormal mark 16 exists on the first sheet 11 and the second sheet 12 that constitute the sheet pair 13 of the bag body 14 reflected in the image 209. Make a judgment based on At this time, the determination unit 101 performs the above determination using a determination model 101a constructed (in other words, generated) in advance by machine learning. Details of the determination model 101a will be described later.

Both the acquisition unit 102 and the determination unit 101 are realized by executing a program using a circuit and a memory connected to the circuit. Therefore, the acquisition unit 102 and the determination unit 101 can also be implemented as an inspection program for causing a computer to execute the functions of the acquisition unit 102 and the determination unit 101 . The acquisition unit 102 and the determination unit 101 may be implemented using separate circuits and memories, respectively, or one circuit and memory may be shared.

Here, the imaging device 200 used to capture the image 209 and the captured image 209 will be described. FIG. 5 is a diagram for explaining an imaging device according to an embodiment. Also, FIG. 6 is a diagram for explaining the content of an image acquired in the embodiment. FIG. 5 is a block diagram showing various components related to imaging by the imaging device 200 and various images output by imaging.

As shown in FIG. 5 and as described above, the imaging device 200 comprises a first camera 201 , a second camera 202 and a third camera 203 . The imaging device 200 is arranged on the second sheet 12 side, and images the sheet pair 13 from the second sheet 12 side. Further, imaging by the imaging device 200 is performed using a first light source 207 provided on the first sheet 11 side and a second light source 208 provided on the second sheet 12 side.

The first camera 201 is a line camera composed of a plurality of imaging elements arranged in a direction perpendicular to the transport direction of the bag 14 along the transport path 400 . By configuring in this way, the first camera 201 can image the entire surface of the conveyed bag body 14 with a small number of imaging elements. As shown in FIG. 6, the first image 204 output by the first camera 201 includes a first grayscale image captured when the light emitted by the first light source 207 has a first luminance, and a first grayscale image. and a second gray-scale image captured when the light has a second luminance different from the first luminance.

For example, when a foreign substance having a light-shielding property is mixed, a foreign substance mark corresponding to the outer shape of the foreign substance is similarly formed at the position where the foreign substance is mixed in both the first grayscale image and the second grayscale image. Further, by generating a differential image between the first grayscale image and the second grayscale image, the influence of wrinkles and the like on the sheet pair 13 can be reduced.

The second camera 202 is a line camera composed of a plurality of imaging elements arranged in a direction perpendicular to the direction in which the bag 14 is conveyed along the conveying path 400 . By configuring in this way, the second camera 202 can image the entire surface of the conveyed bag body 14 with a small number of imaging elements. As shown in FIG. 6 , the second image 205 output by the second camera 202 includes an R image in which the light emitted by the first light source 207 is captured as red light received by the second camera 202 . Also, the second image 205 includes a G image captured as green light in which the light emitted by the first light source 207 is received by the second camera 202 . The second image 205 is output including a B image captured as blue light in which the light emitted by the first light source 207 is received by the second camera 202 . A known method using a bandpass filter or the like is arbitrarily used for such imaging of light of each color.

The R image, the G image, and the B image can emphasize the shading according to the color of the transmitted light. In other words, it is possible to further improve the contrast of a foreign substance or the like of a specific color.

The third camera 203 is a line camera composed of a plurality of imaging elements arranged in a direction perpendicular to the transport direction of the bag 14 along the transport path 400 . By configuring in this way, the third camera 203 can image the entire surface of the conveyed bag body 14 with a small number of imaging elements. As shown in FIG. 6, the third image 206 output by the third camera 203 includes a reflected light image in which the light emitted by the second light source 208 and reflected by the sheet pair 13 is captured. .

The reflected light image is information on the luminance value that depends on the reflection at the interface. is formed.

As described above, in the present embodiment, an example in which the captured image 209 actually includes six types of images captured and output by three types of cameras that capture the same portion of the bag body 14 will be described. However, the image 209 may simply be an image obtained by capturing one type of visible light image.

FIG. 7 is a diagram for explaining a determination model used in the determination unit according to the embodiment; As described above, in the present embodiment, the determining unit 101 performs determination using the determination model 101a, so that at least one determination of whether the sheet pair 13 is good or bad is performed. Such a determination model 101a is a processing module that is included in a program that is executed using the above-described circuit and memory, and that outputs output data determined based on an input image.

As shown in FIG. 7, the judgment model 101a receives an image 209 captured by the imaging device 200 and acquired by the acquisition unit 102 as input data, and outputs a judgment result according to the input. The determination result is output as at least one of good or bad and bad or bad. The determination model 101a is generated in advance by machine learning, and the operation and the like related to the generation of the determination model will be described later with reference to FIGS. 9 and 10. FIG.

[motion]
Next, the operation of the bag making system will be described using FIG. 8 while referring to FIGS. 1 to 5 as appropriate. FIG. 8 is a flow chart showing the operation of the bag making system according to the embodiment.

In the bag-making system of the present embodiment, the bag body 14 is first formed (formation step S11). For example, in the sheet pair 13 in which the first sheet 11 and the second sheet 12 are stacked, upstream of the conveying path 400 shown in FIG. The bag body 14 is formed by

Formation of the bag body 14 is performed, as described above, by stacking the first sheet 11 and the second sheet 12 and joining the ends thereof. Such bonding may be adhesive bonding using an adhesive material, or welding bonding in which the ends are melted by an ultrasonic generator, a heater, or the like to diffuse each other, and then cooled and hardened. good too. An appropriate bonding method may be adopted depending on the material, shape, etc. of the first sheet 11 and the second sheet 12 .

In the bag making system, the sheet pair 13 of the first sheet 11 and the second sheet 12 forming the formed bag body 14 is then inspected by the inspection device 100 (inspection step S12). In the inspection step S12, first, the imaging device 200 captures an image of the sheet pair 13, and the acquisition unit 102 acquires the image (acquisition step S21). In the inspection step S12, at least one of the sheet pair 13 appearing in the image is judged to be good or defective based on the obtained image (judgment step S22).

This determination may be made by determining at least one of whether the sheet pair 13 is good and bad. A good/bad decision is made to determine both whether there is or not. Note that the judgment model 101a generated in advance is used for the judgment of the sheet pair 13 in the judgment step S22.

In the inspection step S12, next, the determination result in the determination step S22 is output (output step S23).

In this manner, the bag making system is controlled based on the determination result output in the inspection step S12. Specifically, the sheet pair 13 to be controlled is determined by determining whether the sheet pair 13 is non-defective or defective in the determination result (determination step S13). If the determination result of the target sheet pair 13 (that is, the sheet pair 13 that appears in the image) is not acceptable or is determined to be defective (Yes in decision step S13), the bag making system is controlled to An operation is performed to enable identification of the bag 14 corresponding to the sheet pair 13 (control step S14).

For example, in the control step S14, the bag-making system processes sheets judged to be unacceptable or defective by attaching stickers, markings, pinholes, or the like to the bag body 14 corresponding to the target sheet pair 13. Make pair 13 identifiable. In other words, the bag making system controls a device that performs operations such as attaching stickers, markings, pinholes, or the like to the sheet pair 13 in the control step S14. In other words, the bag making system further includes a device (not shown) that performs operations such as attaching stickers, markings, pinholes, or the like to the sheet pair 13 .

Further, the transport path 400 may have a first branched path and a second branched path (both not shown) branched at a branch point on the downstream side of the inspection apparatus 100 . The bag making system controls the branch point to convey the normal bag body 14 to the first branch path, and in control step S14, the bag body corresponding to the sheet pair 13 determined to be unacceptable or defective. 14 to the second fork.

By doing so, the bags 14 corresponding to the sheet pairs 13 automatically determined to be non-defective or defective can be collected downstream of the second branch path. Further, for example, if four branch paths are provided, the bags 14 can be sorted according to the four determination results that the sheet pair 13 is good, not defective, not good, and defective. can also In this example, the bag 14 whose sheet pair 13 is determined to be defective is discarded, and the bag 14 whose sheet pair 13 is determined not to be defective or not good is collected and re-inspected (for example, visually by an inspector). inspection), etc.

If the determination result of the target sheet pair 13 is determined to be good or not defective (No in determination step S13), the operation for the sheet pair 13 is terminated. Since this embodiment relates to a bag-making system in which bag bodies 14 are continuously supplied, the bag-making system shifts to the operation of the next transported bag body 14 on the sheet pair 13 (formation step return to S11).

[Generation of decision model]
Next, the configuration and operation for generating the judgment model 101a will be described with reference to FIGS. 9 and 10. FIG. FIG. 9 is a block diagram showing a generation device and an inspection device according to the embodiment. FIG. 9 shows the block diagram of the inspection apparatus 100 shown in FIG. 4 together with a generation device 300 used to generate the judgment model 101a. In the following description of the configuration for generating the determination model 101a, the description of the configuration already described will be omitted or simplified.

As shown in FIG. 9, the determination model 101a used by the inspection device 100 in this embodiment is generated by the generation device 300. FIG. For example, the inspection apparatus maker that manufactures the inspection apparatus 100 owns the generation apparatus 300 and stores the determination model 101a generated in advance by machine learning in the memory of the determination unit 101 or the like. Can be used alone. In other words, the inspection device 100 and the generating device 300 may be realized as separate bodies, and are used while being communicably connected to each other when generating the judgment model 101a. Moreover, when the user who uses the inspection apparatus 100 constructs the determination model 101a according to the application, the inspection apparatus 100 and the generation apparatus 300 may be integrated.

The generation device 300 is a device for generating the determination model 101a used in the determination unit 101 as described above. The generation device 300 specifically includes a reception unit 302 and a generation unit 301 . The receiving unit 302 is a processing unit that receives a teacher image, which is teacher data for generating the determination model 101a by machine learning. As the teacher image, an image of a pair of non-defective sheets, which is determined to be good in the inspection, captured from the second sheet 12 side is used. The reception unit 302 performs processing such as converting the received teacher image into a format that can be processed by the generation unit 301 described later, if necessary.

The generation unit 301 is a processing unit that uses a training image as training data to generate the judgment model 101a by machine learning. The generating unit 301 generates an optimized model as the judgment model 101a by learning, for example, a neural network having a plurality of hierarchies. The neural network generated by the generation unit 301 is generated, for example, by performing machine learning using a teacher image by deep learning. In such machine learning, a technique such as backpropagation corrects the connection counts between artificial neurons in a neural network. An input teacher image is decomposed into luminance values and the like in pixels of the six types of images described above, and each is input to the input layer. In addition, the value to be output is selected from either one of the two values of "good" or "not good" or the two values of "not bad" or "bad". A single value.

Both the receiving unit 302 and the generating unit 301 are realized by executing a program using a circuit and a memory connected to the circuit. Therefore, it is also possible to implement the functions of the reception unit 302 and the generation unit 301 as a program for causing a computer to execute the functions. The reception unit 302 and the generation unit 301 may be implemented using different circuits and memories, respectively, or may be implemented by sharing one circuit and memory. Alternatively, the acquisition unit 102, the determination unit 101, the reception unit 302, and the generation unit 301 may use one common circuit and memory.

The operation of generating the determination model 101a by the generation device 300 will be described below. FIG. 10 is a flow chart showing the operation of the generator. In FIG. 10, since the imaging device 200 is connected via the inspection device 100 as shown in FIG. 9, it is possible to capture the image of the actual good sheet pair and acquire the teacher image.

First, a pair of non-defective sheets determined to be non-defective are supplied (supply step S31). The non-defective sheet pair is supplied by being placed and conveyed on the conveying path 400 in the same manner as the sheet pair 13 in the inspection using the inspection apparatus 100, for example. A good sheet pair fed through the conveying path 400 is imaged by the imaging device 200 , and the captured teacher image is transmitted to the generating device 300 . In the generation device 300, the reception unit 302 receives the teacher image (reception step S32).

The generation device 300 generates the determination model 101a by having a model such as a neural network learn by machine learning using the received teacher image as teacher data. In this manner, by automatically supplying pairs of non-defective sheets one after another and performing machine learning on the determination model 101a, the accuracy of determination by the determination model 101a can be further improved. It should be noted that it is not necessary to use a substantial good sheet pair. For example, if it is possible to obtain a teacher image in which a good sheet pair is captured, the supply step S31 is not essential. That is, the operation of generating the determination model 101a can also be started from the receiving step S32.

The determination model 101a generated in this manner is constructed so as to allow the first pattern 11b formed on the first sheet 11 to be seen through the second sheet 12. FIG. Therefore, the determination model 101a does not determine that the sheet pair 13 corresponding to the image 209 including the projection pattern on which the first pattern 11b is projected is not acceptable or defective. Also, the generated judgment model 101a is constructed so as to allow the perforations 15 formed on the first sheet 11 and the second sheet 12 . Therefore, the judgment model 101a does not judge the sheet pair 13 corresponding to the image 209 including the perforation pattern on which the perforations 15 are projected as non-defective or defective. Also, the generated judgment model 101a is constructed so as to allow the arrangement positions of the first seat 11 and the second seat 12 . Therefore, the judgment model 101a determines that the sheet pair 13 corresponding to the image 209 including the superimposed pattern 14a projected by superimposing the first pattern 11b and the second pattern 12b and the pseudo-abnormal pattern 14b is not acceptable or defective. not determined to be

On the other hand, the generated determination model 101a is constructed to detect scratches, fisheyes, and pinholes formed in the first sheet 11 and the second sheet 12, as well as mixed foreign matter, etc., as abnormalities. . Since these are not included in the non-defective sheet pair, the judgment model 101a is based on an abnormality that appears due to at least one of formed scratches, fisheyes, pinholes, mixed foreign matter, and the like. The sheet pair 13 corresponding to the image 209 containing the marks 16 is determined to be unacceptable or defective.

Since the determination model 101a is optimized in this way, the sheet pair 13 in which the first sheet 11 having the first pattern 11b and the second sheet 12 at least partially translucent are stacked, , the sheet pair 13 formed by joining the ends 11a and 12a of the first sheet 11 and the second sheet 12 to form the bag 14 can be inspected. Thus, the present disclosure allows for more accurate inspection of such sheet pairs 13 .

Instead of the good sheet pair, the judgment model 101a uses a defective sheet pair that is judged to be defective among good and defective sheets, and the sheet pair 13 that is judged to be non-good or defective. May be constructed to disallow.

[Modification]
A modification of the embodiment will be further described below with reference to FIGS. 11 and 12. FIG. FIG. 11 is an overview diagram showing the appearance of a bag-making system according to a modification of the embodiment. 11 differs from FIG. 1 in that the bag body 14 is formed on the downstream side of the inspection device 100. FIG. That is, in this modification, the inspection device 100 inspects the sheet pair 13 before the bag body 14 is formed. To then form the bag 14, the figure shows a forming apparatus 500 for forming the bag. That is, FIG. 11 shows the constituent elements that make up the bag making system 900 .

As shown in FIG. 11 , in this modified example, a first sheet 11 and a second sheet 12 are stacked to form a sheet pair 13 and arranged below the imaging device 200 . The first sheet 11 and the second sheet 12 have a continuous film shape up to the forming device 500 . The configurations of the inspection apparatus 100 and the imaging apparatus 200 are the same as those of the above-described embodiment, and thus description thereof is omitted. The forming apparatus 500 fuses a sheet pair of a predetermined length from the continuous film-shaped sheet pair 13 and joins the ends of the first sheet 11 and the second sheet 12 to each other. An apparatus for forming a bag 14. FIG. The shape of the formed bag body 14 is the same as that of the above embodiment.

Due to the configuration described above, the order of operations of the bag-making system 900 in this modified example is different from that of the bag-making system in the embodiment described with reference to FIG. FIG. 12 is a flow chart showing the operation of the bag making system according to the modification of the embodiment. Specifically, in this modification, the formation step S11 in FIG. 8 is replaced with a formation step S11a that is performed after the control step S14, as shown in FIG. Other operations are the same as those in the above embodiment, so description thereof will be omitted.

A portion of the sheet pair 13 determined to be non-defective or defective by the inspection device 100 is cut off by a fusing mechanism of the forming device 500 so that the portion is not included when forming the bag body 14, for example. and discarded. By doing so, the amount of sheet pairs 13 to be discarded can be reduced, and the effect of reducing the environmental load and cost can be expected.

However, in this configuration, since the inspection is performed before joining the first sheet 11 and the second sheet 12, which is the final step of forming the bag 14, the sheet pair 13 in the formed bag 14 is good or not. At least one determination of failure is not performed. That is, the quality of the bag body 14 itself is not guaranteed. As an example, the application of this configuration may be considered when the product is used under specific conditions such as being inspected again after the product is enclosed in the bag 14 .

[Effects, etc.]
As described above, the inspection apparatus 100 according to the present embodiment includes the sheet pair 13 in which the first sheet 11 having the first pattern 11b and the second sheet 12 at least partially translucent are stacked. The inspection device 100 inspects the sheet pair 13 forming the bag body 14 by joining the ends 11a and 11b of the first sheet 11 and the second sheet 12, respectively. a pair of non-defective sheets determined to be good out of good or defective in the inspection, and a pair of non-defective sheets determined to be defective out of good or defective A determination model 101a constructed by machine learning using a teacher image in which at least one of the pair of defective sheets is picked up from the side of the second sheet 12 determines whether the sheet pair 13 corresponding to the acquired image 209 is good or bad. and a determination unit 101 that determines at least one of

Such an inspection apparatus 100 is used for the case where the first pattern 11b has a pattern that cannot appear on the first sheet 11, which is formed by being transmitted through a translucent portion of the second sheet 12 and being overlapped. can also determine at least one of good and bad. For example, when the first pattern 11b is viewed from the side of the second sheet 12, the first pattern 11b appears as a pattern corresponding to the front and back sides of the sheet. be. Therefore, by using the judgment model 101a that is learned by machine learning the image 209 that can include the first pattern 11b that can be seen from the back side, the first pattern 11b and the first pattern 11b are judged to be non-defective or defective. Therefore, it is possible to accurately determine whether the sheet pair 13 is good or defective. Therefore, the inspection apparatus 100 can accurately inspect the sheet pair 13 using accurate determination results.

Also, for example, the first pattern 11b may be preliminarily formed on the first sheet 11 by letters, figures, numbers, symbols, patterns, or a combination of at least two of these.

According to this, at least one of good and bad of the sheet pair 13 including the first sheet 11 having the first pattern 11b preformed by letters, graphics, numbers, symbols, patterns, or a combination of at least two of these Judgment can be made accurately. Therefore, the inspection apparatus 100 can accurately inspect the sheet pair 13 using accurate determination results.

Further, for example, in the determination by the determination model 101a, the sheet pair 13 corresponding to the image 209 including the projected pattern on which the first pattern 11b is projected is not determined to be non-defective or defective. The sheet pair 13 corresponding to the image 209 including the abnormal mark 16 appearing on the image 209 due to at least one of scratches, fisheyes, and pinholes formed in the sheet pair 13 and foreign matter mixed in the sheet pair 13, It may be determined that it is not good or that it is defective.

According to this, when the pattern formed in the image 209 includes a projection pattern in which the first pattern 11b is projected, it can be distinguished from the abnormal mark 16, so such a projection pattern can be classified as good or bad. Does not contribute to at least one decision. In other words, even if the projected pattern is included in the image 209, the sheet pair 13 is not determined to be non-defective or defective. On the other hand, the abnormal mark 16 appearing on the image 209 due to at least one of scratches, fisheyes, and pinholes formed on the sheet pair 13 and foreign matter mixed in the sheet pair 13 is the projection pattern described above. and contributes to the determination of at least one of good and bad. In other words, when such an abnormal mark 16 is included in addition to the projected pattern, the sheet pair 13 can be determined to be non-defective or defective. Therefore, the inspection apparatus 100 can accurately inspect the sheet pair 13 based on the accurate determination result of at least one of the sheet pair 13 being acceptable or defective.

Further, for example, the second sheet 12 has a second pattern 12b different from the first pattern 11b, and in the determination by the determination model 101a, the first pattern 11b and the second pattern 12b are superimposed and projected. The sheet pair 13 corresponding to the image 209 including the pattern 14a may not be determined as non-good or defective.

According to this, when the pattern formed in the image 209 includes the superimposed pattern 14a projected by superimposing the first pattern 11b and the second pattern 12b, it can be distinguished from the abnormal mark 16. Such a superimposed pattern 14a does not contribute to the determination of at least one of good and bad. In other words, even if the superimposed pattern 14a is included in the image 209, the sheet pair 13 is not determined to be non-defective or defective. Therefore, the inspection apparatus 100 can accurately inspect the sheet pair 13 based on the accurate determination result of at least one of the sheet pair 13 being acceptable or defective.

Also, for example, the second pattern 12b may be pre-formed on the second sheet 12 by letters, figures, numbers, symbols, patterns, or a combination of at least two of these.

According to this, at least one of good and bad of the sheet pair 13 including the second sheet 12 having the second pattern 12b pre-formed with letters, graphics, numbers, symbols, patterns, or a combination of at least two of these. Judgment can be made accurately. Therefore, the inspection apparatus 100 can accurately inspect the sheet pair 13 using accurate determination results.

Further, for example, at least one of the first sheet 11 and the second sheet 12 has a perforation 15 for opening the formed bag body 14, and the perforation 15 is projected according to the determination by the determination model 101a. The sheet pair 13 corresponding to the image 209 containing the sewing pattern is not determined to be unacceptable or defective.

According to this, when the pattern formed in the image 209 includes a perforation pattern in which the perforation 15 is projected, it can be distinguished from the abnormal mark 16. Therefore, such a perforation pattern is at least good and bad. It does not contribute to the judgment of one side. In other words, even if the image 209 includes a sewing machine pattern, the sheet pair 13 is not judged to be good or defective. Therefore, the sheet pair 13 can be accurately inspected based on the accurate determination result of at least one of the sheet pair 13 being acceptable or defective.

Further, for example, the inspection apparatus 100 may inspect the sheet pair 13 after the bag body 14 is formed.

According to this, the inspection device 100 can accurately inspect the sheet pair 13 forming the bag 14 supplied through the forming device 500 for forming the bag 14 .

Further, for example, the inspection apparatus 100 may inspect the sheet pair 13 before the bag body 14 is formed.

According to this, the inspection device 100 can accurately inspect the sheet pair 13 supplied to the forming device 500 for forming the bag body 14 .

Further, for example, the image 209 is captured by the imaging device 200 provided on the second sheet 12 side when the light emitted by the first light source 207 provided on the first sheet 11 side has the first luminance. A captured first grayscale image, and a second grayscale image captured by the imaging device 200 when the light emitted by the first light source 207 has a second brightness different from the first brightness. A first image 204 may be included.

According to this, it is possible to construct the determination model 101a that has learned the feature amount that can appear in the first image 204 including the first grayscale image and the second grayscale image. Therefore, using the determination model 101a, the determination unit 101 determines at least one of the sheet pair 13 to be acceptable or defective based on the feature amount of the first image 204 captured in the sheet pair 13 to be inspected. can be done accurately. Therefore, the inspection apparatus 100 can accurately inspect the sheet pair 13 using accurate determination results.

Further, for example, the image 209 is formed by red light, green light, and red light received by the imaging device 200 provided on the second sheet 12 side. and a second image 205 captured as luminance values of each of blue light.

According to this, features that can appear in the second image 205 captured as luminance values of red light (i.e., R image), luminance values of green light (i.e., G image), and luminance values of blue light (i.e., B image) A judgment model 101a that has learned the quantity can be constructed. Therefore, using the determination model 101a, the determination unit 101 determines at least one of the sheet pair 13 to be acceptable or defective based on the feature amount of the second image 205 captured in the sheet pair 13 to be inspected. can be done accurately. Therefore, the inspection apparatus 100 can accurately inspect the sheet pair 13 using accurate determination results.

Further, for example, the image 209 is received by the imaging device 200 provided on the second sheet 12 side after the light emitted by the second light source 208 provided on the second sheet 12 side is reflected by the sheet pair 13 . may include a third image 206 captured by

According to this, it is possible to construct the determination model 101a that has learned the feature amount that can appear in the third image 206, which is the reflected light image. Therefore, using the determination model 101a, the determining unit 101 determines at least one of the sheet pair 13 to be acceptable or defective based on the feature amount of the third image 206 captured in the sheet pair 13 to be inspected. can be done accurately. Therefore, the inspection apparatus 100 can accurately inspect the sheet pair 13 using accurate determination results.

Moreover, in the bag-making system 900 and the like in the present embodiment, in the sheet pair 13 in which the first sheet 11 having the first pattern 11b and the second sheet 12 at least partially translucent are overlapped, the A forming apparatus 500 that forms the bag body 14 by joining the ends 11a and 12a of the first sheet 11 and the second sheet 12, and the inspection apparatus 100 described above.

Such a bag-making system 900 and the like can form a bag body 14 whose quality is ensured by an accurate inspection based on determination of at least one of good and bad of the constituting sheet pair 13 .

Further, the determination model 101a in the present embodiment is a determination model 101a that is included in a program executed by a computer and that outputs output data determined based on the image 209 input as input data. A pair of sheets 13 in which a first sheet 11 having a pattern 11b and a second sheet 12 at least partially translucent are stacked, and edge portions 11a of each of the first sheet 11 and the second sheet 12 and 12a are joined to form the bag body 14, and when an image 209 obtained by imaging the sheet pair 13 from the second sheet 12 side is input as input data, the sheet pair 13 corresponding to the input image 209 is input. At least one of good and bad is determined, and the determination result is output as output data.

Such a judgment model 101a is for the case where the first pattern 11b has a pattern that cannot appear on the first sheet 11, which is formed by being transmitted through a translucent portion of the second sheet 12 and overlapping. can also determine at least one of good and bad. By using the judgment model 101a which is learned by machine learning the image 209, the first pattern 11b is distinguished from the portion judged to be not good or defective, and at least the good and bad of the sheet pair 13 are determined. One determination can be made accurately.

In addition, the generating device 300 in the present embodiment is a pair of sheets 13 in which the first sheet 11 having the first pattern 11b and the second sheet 12 at least partially translucent are stacked. Based on an image 209 of the sheet pair 13 that forms the bag 14 by joining the ends 11a and 12a of the first sheet 11 and the second sheet 12, at least one of good and bad of the sheet pair 13 is determined. A generating device 300 for generating a judgment model 101a for making a judgment, which includes a pair of non-defective sheets that are judged to be good among good or defective sheets and a defective sheet pair that is judged to be defective among good or defective sheets in the inspection. A reception unit 302 that receives a teacher image obtained by imaging at least one of the non-defective sheet pair from the second sheet 12 side, and a generation unit that generates the determination model 101a by machine learning using the teacher image received by the reception unit 302 as teacher data. 301 and.

Such a generation device 300 can generate the determination model 101a that has the above effects.

Further, the inspection method according to the present embodiment relates to the sheet pair 13 in which the first sheet 11 having the first pattern 11b and the second sheet 12 at least partially translucent are stacked. An inspection method for inspecting a sheet pair 13 forming a bag body 14 by joining the ends 11a and 12a of the sheet 11 and the second sheet 12, wherein the sheet pair 13 is imaged from the second sheet 12 side. an acquisition step S21 for acquiring the image 209 obtained by the inspection, and at least a pair of non-defective sheets determined to be good out of good or defective in the inspection and a pair of defective sheets determined to be defective out of good or defective. Determination of at least one of good and bad of the sheet pair 13 corresponding to the acquired image 209 by the determination model 101a constructed by machine learning using a teacher image, one of which is captured from the second sheet 12 side. and step S22.

Such an inspection method can also be applied to the case where the first pattern 11b has a pattern that cannot appear on the first sheet 11, which is formed by being transmitted through a translucent portion of the second sheet 12 and overlapping. , at least one of good and bad can be determined. By using the judgment model 101a which is learned by machine learning the image 209, the first pattern 11b is distinguished from the portion judged to be not good or defective, and at least the good and bad of the sheet pair 13 are determined. One determination can be made accurately. Therefore, the sheet pair 13 can be accurately inspected using the accurate determination results obtained by the above inspection method.

In the bag-making method of the present embodiment, in the sheet pair 13 in which the first sheet 11 having the first pattern 11b and the second sheet 12 at least partially translucent are stacked, the first sheet Forming step S11 or S11a of forming bag body 14 by joining ends 11a and 12a of 11 and second sheet 12, and determining at least one of good and bad of sheet pair 13 by the above inspection method. and an inspection step S12 for inspecting the sheet pair 13 by.

Such a bag-making method can form a bag body 14 whose quality is ensured by an accurate inspection based on determination of at least one of good and bad of the constituting sheet pair 13 .

Further, the generation method in the present embodiment is the sheet pair 13 in which the first sheet 11 having the first pattern 11b and the second sheet 12 at least partially translucent are stacked, and the first sheet 13 Based on an image 209 of the sheet pair 13 that forms the bag body 14 by joining the ends 11a and 12a of the sheet 11 and the second sheet 12, the sheet pair 13 is inspected to determine whether the sheet pair 13 is good or not. A method for generating a determination model 101a used for determining at least one of defects, in which a pair of non-defective sheets determined to be good out of good or defective in inspection, and a pair of good sheets determined to be defective out of good or defective A reception step S32 for receiving a teacher image of at least one of the pair of defective sheets taken from the second sheet 12 side, and a generation for generating a determination model 101a by machine learning using the teacher image received in the reception step S32 as teacher data. and step S33.

Such a generation method has the same effects as the generation device 300 described above.

(Other embodiments)
As described above, the inspection apparatus and the like according to the present disclosure have been described based on the above embodiments and the like, but the present disclosure is not limited to the above embodiments and the like. For example, a form obtained by applying various modifications that a person skilled in the art can think of for each embodiment, etc., or a form obtained by arbitrarily combining the constituent elements and functions of each embodiment without departing from the scope of the present disclosure. Also included in the present disclosure is the form of

In the above-described embodiment and the like, the configuration for inspecting a pair of sheets in which the first sheet is the vertically lower side and the second sheet is the vertically upper side has been described. may Components constituting such a back surface inspection apparatus may be the same as those of the inspection apparatus, and description thereof will be omitted. A reversing device for reversing the orientation of the pair of sheets is provided between the inspection device and the back surface inspection device. , the posture of the sheet pair is reversed. In this way, the results of the inspection using the image captured from the first sheet side may be used to form bags whose quality is ensured by even more accurate inspection.

Further, in the above-described embodiments and the like, the imaging device has three cameras (first to third cameras) built in one housing. may be placed at different locations. That is, a first image is captured by a first camera at a first position on the transport path, a second image is captured by a second camera at a second position on the transport path, and a third camera is captured at a third position on the transport path. You may image a 3rd image by. It is possible to suppress an increase in size due to the use of a single housing for the imaging device, and to reduce spatial restrictions on installation of the imaging device. Therefore, in such a variant, the applicability of applying the inspection device can be expanded.

Moreover, in the above-described embodiment, the inspection device has been described as being installed in the transport path, but the configuration is not limited to this. The present disclosure also includes placing the sheet pair 13 on a stationary table, such as an inspection stage, and performing an inspection using an image captured by an area camera instead of a line camera.

In addition, the sheet pair described in the above embodiment has a configuration in which two sheets are stacked, but when three or more sheets are stacked and joined to form a three-dimensional bag body, three sheets A sheet pair composed of the above sheets can also be inspected.

Further, for example, the present disclosure can be realized not only as hardware such as an inspection device, but also as a program that includes steps performed by each component such as an inspection device, and a computer-readable recording medium that records the program. It can also be realized. The program may be pre-recorded on a recording medium, or may be supplied to the recording medium via a wide area network including the Internet.

That is, any of the general or specific aspects described above may be implemented in a system, device, integrated circuit, computer program or computer readable recording medium, and any of the system, device, integrated circuit, computer program and recording medium may be implemented. may be implemented in any combination.

The inspection apparatus and the like of the present disclosure are used for the purpose of more accurately judging and inspecting the quality of a pair of sheets forming a bag or the like.

11 1st sheet 11a 1st end 11b 1st pattern 12 2nd sheet 12a 2nd end 12b 2nd pattern 13 sheet pair 14 bag body 14a superimposed pattern 14b pseudo abnormal pattern 15 perforation 16 abnormal mark 100 inspection device 101 determination Unit 101a Judgment model 102 Acquisition unit 200 Imaging device 201 First camera 202 Second camera 203 Third camera 204 First image 205 Second image 206 Third image 207 First light source 208 Second light source 209 Image 300 Generation device 301 Generation unit 302 Receiving Unit 400 Conveyance Path 500 Forming Device 900 Bag Making System

Claims (13)

A pair of sheets in which a first sheet having a first pattern and a second sheet having at least a portion of a light-transmitting property are stacked, and edges of each of the first sheet and the second sheet are joined An inspection device for inspecting a pair of sheets forming a bag by
an acquisition unit that acquires an image of the sheet pair captured from the second sheet side;
At least one of a non-defective sheet pair determined to be good out of good or defective in the inspection and a defective sheet pair determined to be defective out of good or defective is imaged from the second sheet side. a judgment unit that judges at least one of the pair of sheets corresponding to the acquired image as being good or defective by a judgment model constructed by machine learning using the teacher image ,
In the judgment by the judgment model,
caused by at least one of a pseudo-abnormal pattern formed from a projected pattern obtained by projecting the first pattern, scratches, fisheyes, and pinholes formed in the pair of sheets, and foreign matter mixed in the pair of sheets. The pair of sheets corresponding to the image including the abnormal mark is determined to be non-defective or defective by distinguishing the abnormal mark appearing on the image using the
inspection equipment. The inspection apparatus according to claim 1, wherein the first pattern is pre-formed on the first sheet by letters, figures, numbers, symbols, patterns, or a combination of at least two of these. The second sheet has a second pattern different from the first pattern,
In the determination by the determination model, the abnormal trace is determined by distinguishing the abnormal trace from the pseudo-abnormal pattern formed from the superimposed pattern in which the first pattern and the second pattern are superimposed and projected. determining that the pair of sheets corresponding to the image containing the sheet is not good or defective
The inspection device according to claim 1 or 2 . 4. The inspection apparatus according to claim 3 , wherein the second pattern is pre-formed on the second sheet by letters, figures, numbers, symbols, patterns, or a combination of at least two of these. At least one of the first sheet and the second sheet has perforations for opening the formed bag,
In the determination by the determination model, by distinguishing between the sewing pattern on which the perforations are projected and the abnormal marks, the sheet pair corresponding to the image including the abnormal marks is determined to be non-defective or defective. determine that there is
The inspection device according to any one of claims 1 to 4 . The inspection device according to any one of claims 1 to 5 , wherein the inspection device performs the inspection on the pair of sheets after the bag body is formed. The inspection device according to any one of claims 1 to 5 , wherein the inspection device performs the inspection on the sheet pair before the bag body is formed. The image has a first gradation imaged by an imaging device provided on the second sheet side when light emitted by a first light source provided on the first sheet side has a first luminance. and a second grayscale image captured by the imaging device when the light emitted by the first light source is light of a second brightness different from the first brightness. The inspection device according to any one of claims 1 to 7 . The image is composed of red light, green light, and blue light emitted by a first light source provided on the first sheet side and received by an imaging device provided on the second sheet side. The inspection device according to any one of claims 1 to 8, comprising a second image captured as a luminance value. The image is captured by light emitted by a second light source provided on the second sheet side, reflected by the sheet pair, and received by an imaging device provided on the second sheet side. The inspection apparatus according to any one of claims 1 to 9, comprising three images. In a pair of sheets in which a first sheet having a first pattern and a second sheet having at least a portion thereof translucent are stacked, the ends of the first sheet and the second sheet are joined together. a forming device for forming a bag;
A bag making system comprising the inspection device according to any one of claims 1 to 10 . A determination model that is included in a program executed by a computer and outputs output data determined based on an image input as input data,
A pair of sheets in which a first sheet having a first pattern and a second sheet having at least a portion of a light-transmitting property are stacked, and edges of each of the first sheet and the second sheet are joined When an image of the pair of sheets forming the bag is captured from the side of the second sheet as the input data, at least one of good and bad of the pair of sheets corresponding to the acquired image is determined. at least one of a pseudo-abnormal pattern formed from a projected pattern obtained by projecting the first pattern, scratches, fisheyes, and pinholes formed on the pair of sheets, and foreign matter mixed in the pair of sheets. by distinguishing the sheet pair corresponding to the image including the abnormal trace from the abnormal trace appearing on the image due to one reason, determining that the sheet pair corresponding to the image including the abnormal trace is not good or defective; A judgment model that outputs judgment results as the output data. A pair of sheets in which a first sheet having a first pattern and a second sheet having at least a portion of a light-transmitting property are stacked, and edges of each of the first sheet and the second sheet are joined An inspection method for inspecting a pair of sheets forming a bag by
an acquisition step of acquiring an image of the sheet pair captured from the second sheet side;
At least one of a non-defective sheet pair determined to be good out of good or defective in the inspection and a defective sheet pair determined to be defective out of good or defective is imaged from the second sheet side. a determination step of determining at least one of good and bad of the sheet pair corresponding to the acquired image by a determination model constructed by machine learning using the teacher image ,
In the determination step,
caused by at least one of a pseudo-abnormal pattern formed from a projected pattern obtained by projecting the first pattern, scratches, fisheyes, and pinholes formed in the pair of sheets, and foreign matter mixed in the pair of sheets. The pair of sheets corresponding to the image including the abnormal mark is determined to be non-defective or defective by distinguishing the abnormal mark appearing on the image using the
Inspection method.

Images