JP7809803B2 - Inspection device, mounting device, and inspection method - Google Patents

Description

This specification discloses an inspection device, an implementation device, and an inspection method.

Conventionally, an inspection device has been proposed that uses images of a board to inspect the presence or absence of components mounted on it (see, for example, Patent Document 1). This inspection device acquires the brightness value of each pixel from a pre-mounting image, taken before the components are mounted, and generates a histogram showing the number of pixels for each brightness value to set a threshold. Then, for each pixel in the pre-mounting image, pixels with brightness values below the threshold are identified as the board, and pixels with brightness values above the threshold are identified as other than the board, thereby generating a mask image that masks everything other than the board. This mask image is used to extract a specific region corresponding to the component mounting surface from the difference image between the pre-mounting image and the post-mounting image, and the presence or absence of the component is determined based on the brightness difference in the specific region.

Japanese Patent Application Laid-Open No. 2017-034202

The inspection device described above sets a threshold value at a brightness value that is a predetermined brightness higher than the peak value of the histogram, and uses that threshold value to distinguish between boards and non-boards. In this way, it is possible to determine the presence or absence of components by using a histogram of feature values such as brightness to distinguish between components and non-components. However, if brightness varies greatly due to the way the light hits and reflects the components and surrounding background when capturing the image, multiple similar peak values may appear, which could result in an incorrect determination of the presence or absence of components.

The primary purpose of this disclosure is to more accurately determine the presence or absence of an object using an image.

This disclosure takes the following measures to achieve the above-mentioned main objective.

The inspection device of the present disclosure includes:
An inspection device that inspects the presence or absence of an object by image processing,
a setting unit that creates or acquires a frequency distribution in which the frequency of feature values at each pixel of the image is tallied by whether or not the object is present, by machine learning using an image in which the presence or absence of an object is associated, and sets a part of each class of the frequency distribution as a class to be determined;
a determination unit that acquires an image to be inspected, and determines the presence or absence of an object in the image to be inspected by comparing the frequency of feature values belonging to the determination target class among feature values of each pixel of the image to the frequency distribution for each presence or absence of an object belonging to the determination target class;
The gist of the project is to provide the following:

The inspection device disclosed herein uses machine learning to create or acquire a frequency distribution that aggregates the frequency of feature values for each pixel of an image by the presence or absence of an object, and sets a portion of each class in the frequency distribution as the judgment target class. The presence or absence of an object in the target image is then determined by comparing the frequency of feature values belonging to the judgment target class among the feature values for each pixel of the image to be inspected with the frequency of the judgment target class in the frequency distribution by the presence or absence of an object. In this way, by using only some of the judgment target classes rather than all of the classes in the frequency distribution aggregated by the presence or absence of an object, it is possible to narrow the judgment to classes where the presence or absence of an object has a significant impact on the frequency, preventing erroneous judgments. This allows for more appropriate judgment of the presence or absence of an object using an image.

FIG. 1 is a diagram showing an outline of the configuration of a mounting system 10. FIG. 2 is a diagram showing an outline of the configuration of a mounting apparatus 20. FIG. 2 is an explanatory diagram showing an example of a tray T for supplying solder balls SB. FIG. 3 is an explanatory diagram showing an example of a suction nozzle 33 that suctions a solder ball SB. 10 is a flowchart showing an example of a process for setting a class to be judged. FIG. 10 is an explanatory diagram showing an example of an image without parts. FIG. 10 is an explanatory diagram showing an example of a luminance histogram without components. FIG. 10 is an explanatory diagram showing an example of an image with a component. FIG. 10 is an explanatory diagram showing an example of a luminance histogram with components present; FIG. 10 is an explanatory diagram showing an example of a class JC to be judged. 10 is a flowchart showing an example of a part presence/absence determination process. An explanatory diagram showing an example of a judgment in the judgment class JC. 10 is a flowchart showing a modified example of a process for setting a class to be judged.

Next, an embodiment of the present disclosure will be described with reference to the drawings. FIG. 1 is a diagram showing an outline of the configuration of a mounting system 10. FIG. 2 is a diagram showing an outline of the configuration of a mounting device 20. FIG. 3 is an explanatory diagram showing an example of a tray T that supplies solder balls SB. FIG. 4 is an explanatory diagram showing an example of a suction nozzle 33 that suctions solder balls SB. In this embodiment, the left-right direction (X-axis), front-back direction (Y-axis), and up-down direction (Z-axis) are as shown in FIGS. 1 and 2.

As shown in FIG. 1, the mounting system 10 comprises a printing device 11, a print inspection device 12, a mounting device 20, a mounting inspection device 14, and a management device 18. The printing device 11 is a device that prints a viscous fluid such as solder paste onto a substrate B. The print inspection device 12 is a device that inspects the condition of the solder paste printed on the substrate B. The mounting device 20 is a device that mounts components P onto the substrate B. The mounting inspection device 14 is a device that inspects the condition of the components P mounted on the substrate B. Note that the mounting object is assumed to be a flat substrate B, but is not limited to this and may be a three-dimensional base material, etc.

As shown in Figure 2, the mounting device 20 comprises a board processing unit 21, a component supply unit 22, a control device 25, and a mounting unit 30. In addition to performing a mounting process to place components P on a board B, the mounting device 20 also has a function to perform an inspection process to inspect the board B on which components P have been mounted. The board processing unit 21 is a unit that loads, transports, fixes at the mounting position, and transports the board B. The board processing unit 21 comprises a pair of conveyor belts that are spaced apart from each other at the front and back of the board B in Figure 2 and span the left-right direction, and transports the board B by driving the conveyor belts.

The component supply unit 22 includes multiple tape feeders 22a and a tray feeder 22b, and is located at the front of the mounting device 20. The tape feeders 22a are arranged side by side in the left-right direction (X-axis direction) and supply components by feeding a tape from a reel in the front-to-back direction (Y-axis direction), with components housed in multiple recesses formed at predetermined intervals in the longitudinal direction. The tray feeder 22b supplies components by feeding a tray T housing components in the front-to-back direction (Y-axis direction). As shown in Figure 3, the top surface of the tray T houses spherical solder balls SB as components P in multiple recesses C formed, for example, in a grid pattern. The tray feeder 22b feeds this tray T and supplies the solder balls SB.

The mounting unit 30 is a unit that picks up components P from the component supply unit 22 and places them on a board B fixed to the board processing unit 21. The mounting unit 30 includes a head moving unit 31, a mounting head 32, a suction nozzle 33, and a mark camera 34. The head moving unit 31 includes a slider that moves in the X and Y directions along a guide rail, and a motor that drives the slider. The mounting head 32 is removably attached to the slider and moved in the X and Y directions by the head moving unit 31. One or more suction nozzles 33 (e.g., 1, 4, 8, 16, etc.) are removably attached to the underside of the mounting head 32. The suction nozzle 33 is a holding member that picks up and holds components P using negative pressure.

The mark camera 34 is disposed on the underside of the mounting head 32 (or slider) so that its imaging range is downward, and moves in the X and Y directions as the mounting head 32 moves. The mark camera 34 captures images of reference marks attached to the board B and components P mounted on the board B from above, and outputs the images to the control device 25.

The mounting device 20 also includes a part camera 23, a nozzle stocker 24, and a display/operation unit 28. The part camera 23 is disposed between the component supply unit 22 and the substrate processing unit 21 so that the upper portion is within the imaging range. The part camera 23 captures an image of the component P when the mounting head 32 holding the component P passes above the part camera 23 and outputs the image to the control device 25. The nozzle stocker 24 stocks multiple types of suction nozzles 33 and is disposed between the component supply unit 22 and the substrate processing unit 21. The suction nozzles 33 of the mounting head 32 are replaced as appropriate depending on the type and size of the component P. For example, if the mounting target is a solder ball SB, the nozzle 33 is replaced with a suction nozzle 33 for solder ball SB (see Figure 4). This suction nozzle 33 has multiple (16 in Figure 4) suction ports 33a arranged in a grid pattern, and each suction port 33a is used to suction a solder ball SB in a recess C of the tray T.

The display operation unit 28 is provided at the front of the mounting device 20 (see Figure 1) and includes, for example, a touch panel and operation buttons. This display operation unit 28 displays various images and information to the worker and allows the worker to input various operation instructions, such as selection operations.

The control device 25 is configured as a microprocessor centered on the CPU 26 and includes a memory unit 27 for storing various data. In addition to controlling the entire mounting device 20, the control device 25 also inspects whether the suction nozzle 33 is picking up a component P, whether the shape of the component P is within an acceptable range, and the mounting state (mounted or not) of the component P on the board B. The control device 25 outputs control signals to the board processing unit 21, the component supply unit 22, the part camera 23, the display operation unit 28, and the mounting unit 30, and inputs signals from the board processing unit 21, the component supply unit 22, the part camera 23, the display operation unit 28, and the mounting unit 30. The memory unit 27 stores information regarding the size and shape of the component P, the mounting order for the component P on the board B, the placement position of the component P, and the type of suction nozzle 33 capable of picking up the component P.

The management device 18 is a computer that manages information about each device in the mounting system 10. The management device 18 comprises a control unit, a memory unit, a display device, and an input device. The control unit is configured as a microprocessor centered around a CPU. The memory unit stores information that manages production in the mounting system 10 and information about the various processes performed by each device.

The following describes the operation of the mounting device 20 configured in this manner, illustrating the mounting process of solder balls SB as components P and the post-mounting inspection process. During the mounting process, the CPU 26 of the control device 25 controls the head moving unit 31 to move the mounting head 32, equipped with the suction nozzle 33 for the solder balls SB, above the component supply position of the tray feeder 22b (tray T). Next, the CPU 26 controls the mounting head 32 to lower the suction nozzle 33 and suction the solder balls SB in the recesses C of the tray T. Next, the CPU 26 controls the head moving unit 31 to move the mounting head 32 above the part camera 23, causing the part camera 23 to capture an image of the solder balls SB suctioned by the suction nozzle 33. The CPU 26 also processes the captured image to determine the positional deviation of each solder ball SB suctioned by the suction nozzle 33 and correct the target mounting position based on the positional deviation. For example, the target mounting position is corrected so that the overall positional deviation of each of the multiple solder balls SB is minimized. Then, the CPU 26 controls the head moving unit 31 so that the mounting head 32 moves above the substrate B, and controls the mounting head 32 so that the suction nozzle 33 is lowered to release the suction, thereby mounting the solder ball SB at the target mounting position on the substrate B.

Furthermore, once the mounting of the solder balls SB is completed, the CPU 26 performs an inspection process to inspect the mounting state. In the inspection process, after the solder balls SB are mounted, the mark camera 34 captures an image from above, and the brightness values of the pixels acquired from the image are compared with brightness values for the presence or absence of solder balls SB in the judgment target class JC, which have been set in advance by machine learning, to inspect the presence or absence of the solder balls SB. Note that after mounting is complete, a state in which the solder balls SB are not on the board B occurs, for example, when the solder balls SB do not release from the suction port 33a of the suction nozzle 33 and remain sucked (take-home component). Below, before explaining the inspection process, a process for setting the judgment target class JC using machine learning will be explained. Figure 5 is a flowchart showing an example of the judgment target class setting process.

In the judgment target class setting process, the CPU 26 first acquires a component-free learning image of a board B on which no solder balls SB are mounted, captured by the mark camera 34 (S100), and acquires the brightness value of each pixel in the inspection area A of the component-free learning image (S110). The CPU 26 may acquire multiple component-free learning images and acquire the brightness value of each pixel from the inspection area A of each image. Next, the CPU 26 tally the brightness values of each acquired pixel for each predetermined brightness class to create a component-free brightness histogram (S120).

Next, the CPU 26 acquires a training image with components present, captured by the mark camera 34 of the board B on which the solder balls SB (components P) are mounted (S130), and acquires the brightness value of each pixel in the inspection area A of the training image with components present (S140). The CPU 26 may acquire multiple training images with components present and acquire the brightness value of each pixel from the inspection area A of each image. The CPU 26 then aggregates the acquired brightness values of each pixel for each predetermined brightness class to create a brightness histogram for the presence of components (S150). In S100 and S130, training images without components and training images with components present may be acquired by capturing images with the mark camera 34 before and after mounting the solder balls SB (components P) on the board B.

Here, FIG. 6 is an explanatory diagram showing an example of an image without components, and FIG. 7 is an explanatory diagram showing an example of a brightness histogram without components. Also, FIG. 8 is an explanatory diagram showing an example of an image with components, and FIG. 9 is an explanatory diagram showing an example of a brightness histogram with components. FIG. 6 shows a training image captured before the solder balls SB were mounted on the board B, and FIG. 8 shows a training image captured after the solder balls SB were mounted on the board B. Each image in this embodiment is a grayscale image, for example, with each pixel having a brightness value of 256 levels (0 to 255). The inspection area A used in the inspection process is indicated by a dotted line (see FIGS. 6 and 8). In the inspection process, the brightness value of each pixel within the inspection area A is used to determine the presence or absence of the solder balls SB. The inspection area A is set to be slightly larger than the solder balls SB shown in the image so that the solder balls SB will fit within the area even if the mounting position varies (see FIG. 8). Therefore, in the inspection area A in the image, in addition to the solder balls SB, the solder paste SP printed (applied) on the upper surface of the substrate B and the surrounding electrodes E and the like are captured as background.

In the component-free image of Figure 6, due to the absence of solder balls SB, many dark luminance values are obtained from the pixels representing board B within inspection area A. Furthermore, slightly brighter luminance values are obtained from the pixels representing solder paste SP within inspection area A. On the other hand, in the component-present image of Figure 8, due to the presence of solder balls SB, many bright luminance values are obtained from the pixels in the center of the solder balls SB, and slightly darker luminance values are obtained from the pixels in the peripheral areas of the solder balls SB than in the central area. Even in the peripheral areas, bright luminance values are obtained in some places due to reflected light from neighboring solder balls SB mounted in a grid pattern. Furthermore, even in the presence of solder balls SB, dark luminance values are obtained from the background outside of the solder balls SB. In a histogram showing the frequency distribution of luminance values, luminance values ranging from 0 to 255 are divided into multiple classes (also called luminance classes), and the frequency of luminance values belonging to each class is tallied. In this embodiment, luminance values are divided into eight luminance classes 1 to 8, each with a value of 32, as an example, but the number of classes is not limited to eight. When there were no components, the learning results showed that the frequency was highest in luminance class 4, with relatively high frequencies in luminance classes 1, 5, and 6, and low frequencies in the other luminance classes, as shown in Figure 7. On the other hand, when there were components, the learning results showed that the frequency was highest in luminance class 8, with relatively high frequencies in luminance classes 5 and 6, and low frequencies in the other classes, as shown in Figure 9. Note that a frequency also appeared in luminance class 1 due to the influence of the background other than the solder balls SB.

The control device 25 selects the class with the largest frequency difference d between the presence and absence of parts in the brightness histogram created in this way (S160). Next, the control device 25 sets the selected class as the class JC to be judged (S170), stores the class JC to be judged and the frequencies N0 and N1 for the presence and absence of parts in the class JC to be judged in the memory unit 27 (S180), and ends this process.

Figure 10 is an explanatory diagram showing an example of a judgment target class JC. In Figure 10, the brightness histograms of the bar graphs shown as examples in Figures 7 and 8 are displayed together as a frequency line graph, with the absence of parts indicated by a dotted line and the presence of parts indicated by a dashed line. Furthermore, the frequency difference d calculated for each brightness class as the difference between the frequency of parts present and the frequency of parts not present is designated as frequency differences d1 to d8, but frequency difference d3 is omitted from the illustration as it is approximately 0. In this example, the frequency difference d8 for brightness class 8 is the largest, and brightness class 8 is set as the judgment target class JC. Furthermore, the frequency N0 of parts not present and the frequency N1 of parts present in brightness class 8 are stored in memory unit 27.

Next, the process for determining whether or not a component is present during inspection will be described. Figure 11 is a flowchart showing an example of the process for determining whether or not a component is present. In the process for determining whether or not a component is present, the CPU 26 first uses the mark camera 34 to capture an image of the board B after mounting the solder balls SB (target components) to obtain an image to be inspected (S200). Next, the CPU 26 obtains the brightness value of each pixel within the inspection area A in the target image, and counts the frequency Nj of pixels whose brightness values belong to the judgment target class JC (S210). In S210, the CPU 26 counts only the frequency of pixels whose brightness values belong to the judgment target class JC, and does not count the frequency of classes other than the judgment target class JC.

Next, the CPU 26 compares the counted frequency Nj with the frequency of the judgment class JC of the learning result, i.e., the frequency NO of no component and the frequency N1 of component presence in the judgment class JC (S220), and determines whether the frequency Nj is closer to the frequency N1 of component presence than the frequency NO of no component (S230). If the CPU 26 determines that the frequency Nj is closer to the frequency N1 of component presence, it determines that the target component is present, i.e., that a solder ball SB has been mounted (S240), and terminates this process. On the other hand, if the CPU 26 determines that the frequency Nj is closer to the frequency NO of no component rather than the frequency N1, it determines that the target component is absent, i.e., that a solder ball SB has not been mounted, and a mounting error has occurred (S250), and terminates this process. Note that if the CPU 26 determines that a mounting error has occurred, it displays this information on the display/operation unit 28 to notify the operator of the error.

Figure 12 is an explanatory diagram showing an example of a judgment in the judgment target class JC. For ease of explanation, Figure 12 shows a frequency line graph of the image to be inspected along with the frequency line graph of the learning result shown in Figure 10. However, in this embodiment, the frequency of pixels whose brightness values belong to classes other than the judgment target class JC is not counted, so a frequency line graph (frequency distribution) is not actually created. As shown in the figure, the frequency Nj of the judgment target class JC is closer to the frequency N1 of component presence than the frequency N0 of component absence. Therefore, in this example, it is judged that the target component is present, i.e., that a solder ball SB has been mounted.

It is also possible to determine the presence or absence of a target component based on the entire brightness histogram, including the frequencies of classes other than the target class JC. However, doing so would make the determination difficult, as the frequency of the target component in brightness classes 2 and 7 in Figure 12 is close to both the frequency of component presence and absence, while the frequency of the target component in brightness class 3 deviates from both the frequency of component presence and absence. Furthermore, in brightness class 5, the frequency of the target component image is closer to the frequency of component absence, which could lead to an erroneous determination. Therefore, in this embodiment, the presence or absence of a solder ball SB is determined using only brightness class 8 of the brightness histogram, which has the greatest influence on the brightness of the solder ball SB. This eliminates the influence of background other than the solder ball SB, preventing erroneous determination and enabling accurate determination of the presence or absence of the solder ball SB.

Here, the correspondence between the components of this embodiment and the components of the present disclosure will be clarified. The control device 25 of this embodiment corresponds to the inspection device of the present disclosure, the CPU 26 that executes the judgment target class setting process corresponds to the setting unit, and the CPU 26 that executes the component presence/absence determination process corresponds to the determination unit. The suction nozzle 33 corresponds to the holding member, the mounting unit 30 (excluding the mark camera 34) corresponds to the mounting unit, and the mark camera 34 corresponds to the imaging unit. Furthermore, by explaining the judgment target class setting process and the component presence/absence determination process, an example of the inspection method of the present disclosure is also clarified.

The control device 25 of the mounting device 20 described above uses machine learning to create a brightness histogram that aggregates the frequency of brightness values according to the presence or absence of solder balls SB (components), and sets a portion of this to the judgment target class JC. The presence or absence of solder balls SB is then determined by comparing the frequency Nj of the brightness values of each pixel in the image to be inspected that belongs to the judgment target class JC with the frequencies N0 and N1 of the judgment target class JC in the brightness histogram. In this way, by using a portion of the judgment target class JC in the brightness histogram, it is possible to narrow down the judgment to classes where the presence or absence of solder balls SB has a significant impact on the frequency, preventing erroneous judgments.

In addition, because the judgment class JC is set based on the frequency difference d between the presence and absence of solder balls SB belonging to each class of the histogram, judgment can be made using only the class where the influence of the presence or absence of solder balls SB is reflected in the frequency difference d. In particular, because the class with the largest frequency difference d is set as the judgment class JC, the presence or absence of solder balls SB can be appropriately judged using only the class where the influence of the presence or absence of solder balls SB is significantly reflected.

In addition, the control device 25 counts the frequency Nj of brightness values belonging to the judgment target class JC from each pixel of the target image, and does not count the frequency of brightness values belonging to other classes. Therefore, since it counts only the frequency Nj of brightness values necessary for judgment and does not count the frequency of unnecessary brightness values, it is possible to quickly determine the presence or absence of solder balls SB.

Furthermore, after mounting the solder balls SB picked up by the suction nozzle 33 of the mounting unit 30 onto the substrate B, the mounting device 20 uses the mark camera 34 to capture an image of the substrate B and determine whether the solder balls SB have been mounted. This allows for efficient mounting of the solder balls SB onto the substrate B and inspection to determine whether they have been mounted reliably.

It goes without saying that the present disclosure is in no way limited to the above-described embodiments, and can be implemented in various forms as long as they fall within the technical scope of the present disclosure.

In the above-described embodiment, one judgment class JC was set where the frequency difference d between the presence and absence of solder balls SB was greatest. However, this is not limited to this. At least one judgment class JC may be set based on the brightness trend of the component presence/absence, i.e., the frequency difference d, and multiple judgment class JCs may also be set. For example, a predetermined number of classes may be set as judgment class JCs in descending order of frequency difference d. In the example of Figure 10, the frequency difference d is d8 and then d4 in descending order, so these two classes 4 and 8 may be set as judgment class JCs. Furthermore, one or more classes in which the frequency difference d is equal to or greater than a predetermined value may be set as judgment class JCs. Using multiple judgment class JCs can further improve the accuracy of component presence/absence determination.

In the embodiment, the frequency Nj of brightness values belonging to the class JC to be judged is counted, and the frequency of brightness values belonging to other classes is not counted, but this is not limited to this, and the frequency of all classes may be counted, and the presence or absence of parts may be judged based only on the frequency of the class JC to be judged.

In the embodiment, the judgment class JC is set based on the frequency difference d, but this is not limited to this and may be done as follows. Figure 13 is a flowchart showing a modified judgment class setting process. In the modified example, the same processes as in the embodiment are assigned the same step numbers and descriptions are omitted. In this process, after creating a brightness histogram, the CPU 26 displays the brightness histogram (e.g., Figure 10) on the display screen of the display operation unit 28, instructs the setting (selection) of the judgment class JC (S162), and waits for a setting instruction to be input (S164). Note that the CPU 26 may display a message on the display screen of the display operation unit 28 prompting the user to set the judgment class JC by selecting one or more (at least some) classes.

The CPU 26 then sets one or more classes based on the input setting instructions as the judgment class JC (S170b). This allows the worker's (user's) intentions and experience to be reflected in the setting of the judgment class JC, making it possible to more appropriately prevent erroneous determination of the presence or absence of solder balls SB (components). The judgment class set by the control device 25 (CPU 26) based on the frequency difference d may be displayed on the display screen so that the worker can change it.

In the embodiment, the present disclosure is applied to determining the presence or absence of solder balls SB, but is not limited to this and may be applied to determining the presence or absence of components P mounted on board B. Alternatively, it is not limited to determining the presence or absence of components P mounted on board B, but may be applied to determining the presence or absence of components P adsorbed by suction nozzles 33. For example, an image captured by parts camera 23 of a solder ball SB adsorbed by a solder ball SB adsorption nozzle 33 may be used as the target image to determine the presence or absence of solder balls SB in each suction port 33a. Furthermore, it is not limited to components P, and may be applied to inspecting the presence or absence of objects other than components P. Furthermore, while the example illustrates determination using the brightness value of each pixel, it is not limited to brightness values, and any feature value obtainable from each pixel may be used.

In the embodiment, the control device 25 of the mounting device 20 inspects for the presence or absence of components, but this is not limited to this. A dedicated inspection device, such as the mounting inspection device 14, may also inspect for the presence or absence of components. Alternatively, an external inspection device, such as the management device 18, may acquire an image of the board B after mounting using the mark camera 34 and inspect for the presence or absence of components.

In the embodiment, the control device 25 performs the judgment class setting process and the part presence/absence determination process. However, both processes do not necessarily have to be performed by the same device, and may be performed by separate devices. For example, an external device such as the management device 18 may create a brightness histogram using machine learning, and the control device 25 may acquire the results and set the judgment class JC and frequencies N0 and N1. Alternatively, the management device 18 may perform the judgment class setting process, and the control device 25 may acquire the results. The part presence/absence determination process may then be performed by the control device 25, as in the embodiment. In this way, the judgment class setting process and the part presence/absence determination process may be shared among multiple devices, and an inspection system equipped with these multiple devices may be used.

The inspection device of the present disclosure may be configured as follows. For example, in the inspection device of the present disclosure, the setting unit may set the judgment target class based on the difference in frequency between the presence and absence of an object belonging to each class of the frequency distribution. This allows judgment to be made using only classes in which the effect of the presence or absence of an object is reflected in the difference in frequency, thereby more appropriately preventing erroneous judgment of the presence or absence of an object.

The inspection device disclosed herein may include a display unit that displays the frequency distribution and an input unit that inputs setting instructions from the user, and the setting unit may set the classification to be judged based on the setting instructions input to the input unit from among the classifications of the frequency distribution displayed on the display unit. This allows the user's intentions and experience to be reflected in the setting of the classification to be judged, thereby more appropriately preventing erroneous judgments about the presence or absence of an object.

In the inspection device disclosed herein, the determination unit may count the frequency of feature values belonging to the determination target class for each pixel of the target image, and not count the frequency of feature values belonging to classes other than the determination target class. In this way, only the frequency of feature values necessary for determination is counted, and unnecessary feature values are not counted, allowing for a quick determination of the presence or absence of an object.

The mounting device disclosed herein comprises any of the inspection devices described above; a mounting unit having a holding member for holding an object and mounting the object held by the holding member onto a board; and an imaging unit that captures an image of the board after the object has been mounted as the target image. Therefore, similar to the inspection device described above, it is possible to more appropriately determine the presence or absence of an object using an image.

The disclosed inspection method is an inspection method for inspecting the presence or absence of an object through image processing, and includes the steps of: (a) using machine learning to create or acquire a frequency distribution in which the frequency of feature values at each pixel of an image is aggregated by the presence or absence of an object through image processing, and setting some of the classes in the frequency distribution as classes to be judged; and (b) acquiring an image to be inspected, and determining the presence or absence of an object in the target image by comparing the frequency of feature values at each pixel of the target image that belong to the class to be judged with the frequency of feature values at each pixel of the target image that belong to the class to be judged for the presence or absence of an object in the frequency distribution. Similar to the inspection device described above, the inspection method disclosed herein can more appropriately judge the presence or absence of an object using an image. This inspection method may employ various aspects of the inspection device described above, or additional steps may be added to realize each function.

This specification also discloses the technical idea of changing "an inspection device described in any one of claims 1 to 3" in claim 5 of the original application to "an inspection device described in any one of claims 1 to 4."

This disclosure can be used in the technical field of inspecting the presence or absence of an object using image processing.

10 Mounting system, 11 Printing device, 12 Printing inspection device, 14 Mounting inspection device, 18 Management device, 20 Mounting device, 21 Board processing unit, 22 Component supply unit, 22a Tape feeder, 22b Tray feeder, 23 Parts camera, 24 Nozzle stocker, 25 Control device, 26 CPU, 27 Memory unit, 28 Display operation unit, 30 Mounting unit, 31 Head movement unit, 32 Mounting head, 33 Suction nozzle, 33a Suction port, 34 Mark camera, A Judgment area, B Board, C Recess, E Electrode, P Component, SB Solder ball, SP Solder paste, T Tray.

Claims (6)

An inspection device that inspects the presence or absence of an object by image processing,
a setting unit that creates or acquires a frequency distribution in which the frequency of feature values at each pixel of the image is tallied by whether or not the object is present, by machine learning using an image in which the presence or absence of an object is associated, and sets a part of each class of the frequency distribution as a class to be determined;
a determination unit that acquires an image to be inspected, and determines the presence or absence of an object in the image to be inspected by comparing the frequency of feature values belonging to the determination target class among feature values of each pixel of the image to the frequency distribution for each presence or absence of an object belonging to the determination target class;
An inspection device comprising: the setting unit sets the determination target class based on a difference in frequency between the presence and absence of an object belonging to each class of the frequency distribution.
The inspection device according to claim 1 . a display unit that displays the frequency distribution;
an input unit for inputting setting instructions from a user;
Equipped with
the setting unit sets the class to be judged from among the classes of the frequency distribution displayed on the display unit based on a setting instruction input to the input unit.
The inspection device according to claim 1 . the determination unit counts the frequency of feature values belonging to the determination target class from each pixel of the target image, and does not count the frequency of feature values belonging to classes other than the determination target class;
4. The inspection device according to claim 1. The inspection device according to any one of claims 1 to 3;
a mounting unit having a holding member for holding an object and for mounting the object held by the holding member on a substrate;
an imaging unit that captures an image of the board after the object has been mounted as the object image;
A mounting device comprising: An inspection method for inspecting the presence or absence of an object by image processing, comprising:
(a) creating or acquiring a frequency distribution in which the frequency of feature values at each pixel of the image is tallied by whether or not the object is present, by machine learning using an image in which the presence or absence of an object is associated, and setting a portion of each class of the frequency distribution as a class to be determined;
(b) acquiring an image to be inspected, and judging the presence or absence of an object in the image to be inspected by comparing the frequency of feature values belonging to the judgment target class among feature values of each pixel of the image to the frequency distribution for the presence or absence of an object belonging to the judgment target class;
An inspection method including: