JP7471207B2 - Inspection device, inspection method, and inspection program - Google Patents

Description

The present invention relates to an inspection device, an inspection method, and an inspection program that inspects multiple objects based on an input image that contains the objects as subjects.

Technology is known that inspects multiple objects based on an input image that contains the objects as subjects. One example of the application of such technology is the quality inspection of multiple laser diode chips formed on a wafer.

For example, Patent Document 1 is an example of a document that discloses such a technology. In the inspection device described in Patent Document 1, first, a low-resolution image with a lower resolution than the input image is generated from an input image that contains multiple objects as subjects. Then, a specific process that identifies a region of interest that contains each object in the input image is performed with reference to the low-resolution image, and a classification process that classifies the objects contained in that region is performed with reference to the input image. This succeeds in reducing the processor load in the specific process without sacrificing the classification accuracy in the classification process.

Patent Publication No. 2020-042754

However, conventional inspection techniques may fail to identify the region of interest that contains the object if a foreign object is attached to the object or if the image of the object in the input image contains optical or electrical noise. For example, when identifying the region of interest using template matching, if the degree of match between the template and the image of the object is reduced due to the above-mentioned foreign object or noise, there is a high possibility that the region of interest that contains the object will not be identified.

One aspect of the present invention has been made in consideration of the above problems, and aims to realize an inspection technique that is less likely than conventional techniques to fail to identify a region of interest that contains an object in an input image.

The inspection device according to aspect 1 of the present invention includes a single or multiple processors that execute a first identification process for identifying multiple regions of interest, each of which includes a single object, in an input image that includes multiple objects as subjects; a second identification process for identifying single or multiple regions of interest, each of which includes a single object, in the input image, that are different from the regions of interest identified in the first identification process, based on the periodicity of the multiple regions of interest identified in the first identification process; and an inspection process for performing image inspection of the objects included in the regions of interest, based on each of the multiple regions of interest identified in the first identification process and the single or multiple regions of interest identified in the second identification process.

In the inspection device according to aspect 2 of the present invention, in addition to the configuration of the inspection device according to aspect 1, the processor executes the first identification process using template matching.

In the inspection device according to aspect 3 of the present invention, in addition to the configuration of the inspection device according to aspect 1 or 2, the processor executes, in the second identification process, (1) a first calculation process that calculates one or both of the interval between representative points of two adjacent attention areas aligned in a first direction among the multiple attention areas identified in the first identification process and the interval between representative points of two adjacent attention areas aligned in a second direction intersecting the first direction among the multiple attention areas identified in the first identification process, and (2) a second calculation process that calculates the coordinates of the representative points of an attention area other than the attention area identified in the first identification process based on the interval calculated in the first calculation process.

In the inspection device according to aspect 4 of the present invention, in addition to the configuration of the inspection device according to any one of aspects 1 to 3, the processor is configured to execute the inspection process using a trained model that receives as input each of the multiple regions of interest identified in the first identification process and the single or multiple regions of interest identified in the second identification process, and outputs the class to which the object contained in the input region of interest belongs.

The inspection method according to aspect 5 of the present invention includes a first identification process in which a single or multiple processors identify multiple regions of interest each including a single object in an input image containing multiple objects as subjects, a second identification process in which the processors identify single or multiple regions of interest each including a single object in the input image, the regions of interest being different from the regions of interest identified in the first identification process, based on the periodicity of the multiple regions of interest identified in the first identification process, and an inspection process in which the processors perform image inspection of the objects included in the regions of interest based on each of the multiple regions of interest identified in the first identification process and the single or multiple regions of interest identified in the second identification process.

The inspection program according to aspect 6 of the present invention is an inspection program for operating a computer as an inspection device according to any one of aspects 1 to 4, and causes a processor included in the computer to execute each of the above processes.

According to one aspect of the present invention, it is possible to realize an inspection technique that is less likely than conventional techniques to fail to identify a region of interest that contains an object in an input image.

1 is a block diagram showing a configuration of an inspection device according to an embodiment of the present invention; FIG. 2 is a flow chart showing the flow of an inspection method according to an embodiment of the present invention. 3 is a flow chart showing a specific example of a specific process included in the inspection method of FIG. 2. FIG. 4 is a diagram illustrating an example of execution of the specific process in FIG. 3 . 3 is a flow diagram showing a specific example of a generation process included in the inspection method of FIG. 2. FIG. 6 illustrates an example of execution of the generation process of FIG. 5 . 3 is a diagram showing an application example of the inspection device of FIG. 1 and the inspection method of FIG. 2;

(Configuration of the inspection device)
The configuration of an inspection device 1 according to an embodiment of the present invention will be described with reference to Fig. 1. Fig. 1 is a block diagram showing the configuration of the inspection device 1.

The inspection device 1 is a device that inspects each object Oi (i=1, 2, ..., n) based on an input image Img that includes a number of periodically arranged objects O1, O2, ..., On (n is a natural number equal to or greater than 2) as subjects. The objects O1, O2, ..., On can be, for example, laser diode chips formed on a wafer.

As shown in FIG. 1, the inspection device 1 includes a memory 11, a processor 12, and a storage 13. The memory 11, the processor 12, and the storage 13 are connected to each other via a bus (not shown). This bus may further be connected to an input/output interface (not shown) and a communication interface (not shown). This input/output interface is used, for example, to input an input image from an external device (e.g., a camera) to the inspection device 1, or to output inspection results from the inspection device 1 to an external device (e.g., a display). In addition, this communication interface is used, for example, for the inspection device 1 to receive an input image provided from an external device (e.g., another computer connected to a camera), or for the inspection device 1 to transmit inspection results to be provided to an external device (e.g., another computer connected to a display).

The memory 11 is configured to record an inspection program P for implementing an inspection method S1 described later, and a trained model M used in the inspection method S1 described later. Note that, for example, a semiconductor random access memory (RAM) or the like can be used as the memory 11. Furthermore, for example, a convolutional neural network (CNN) can be used as the trained model M. Furthermore, a logistic regression model, a support vector machine, a random forest, or the like may also be used as the trained model M.

The processor 12 is configured to execute the inspection method S1 described below in accordance with the inspection program P stored in the memory 11. The processor 12 may be, for example, a CPU (Central Processing Unit), a GPU (Graphic Processing Unit), a microprocessor, a digital signal processor, a microcontroller, an ASIC (Application Specific Integrated Circuit) such as a TPU (Tensor Processing Unit), or a combination of these.

The storage 13 is configured to store (non-volatilely store) the above-mentioned inspection program P and the above-mentioned trained model M. When executing the inspection method S1 described below, the processor 12 expands the inspection program P and trained model M stored in the storage 13 onto the memory 11 and references them. Note that the storage 13 may be, for example, a flash memory, a hard disk drive (HDD), a solid state drive (SSD), or a combination of these.

In this specification, "image" refers to a two-dimensional array of pixel values. A pixel value in a monochrome image is, for example, a single numerical value corresponding to brightness, and a pixel value in a color image is, for example, three or four numerical values corresponding to a format such as RGB, CMY, YUV, etc. Also, an area in an input image, such as the "area of interest" described below, is also a two-dimensional array of pixel values and can be considered an image.

The inspection performed by the inspection device 1 may be a binary judgment that does not distinguish between the types of normality and abnormality, or a multi-value judgment that distinguishes between the types of normality and abnormality. In the case of a binary judgment, the inspection result is one of two classes (normal, abnormal) and is expressed, for example, by a Boolean type numerical value. In this case, the judgment result may be a real number type numerical value or a combination thereof that represents one or both of the probability of normality and the probability of abnormality. In the case of a multi-value judgment, if N types of normality and M types of abnormality can be distinguished, the inspection result is one of N+M classes (normal 1, normal 2, ..., normal M, abnormal 1, abnormal 2, ..., abnormal M), and can be expressed, for example, by an integer type numerical value. In this case, the judgment result may be a real number type numerical value or a combination thereof that represents some or all of the probability of normality 1, the probability of normality 2, ..., the probability of normality N, the probability of abnormality 1, the probability of abnormality 2, ..., and the probability of abnormality M. Here, one of N and M is a natural number of 1 or more, and the other is a natural number of 2 or more. Furthermore, the test results may include results that are neither normal nor abnormal (or the probability of them being so), such as "pending judgment."

Note that, although the configuration in which inspection method S1 is executed by a single processor 12 provided in a single computer has been described here, the present invention is not limited to this. In other words, it is also possible to adopt a configuration in which inspection method S1 is executed jointly by multiple processors provided in a single computer, or distributed across multiple computers.

In addition, although a configuration in which the trained model M is stored in a single memory 11 provided in a single computer has been described here, the present invention is not limited to this. In other words, it is also possible to adopt a configuration in which the trained model M is stored in a single computer or in multiple memories provided in a distributed manner in multiple computers.

The inspection program P for causing the processor 12 to execute the inspection method S1 may be recorded on a computer-readable, non-transitory, tangible recording medium. This recording medium may be the memory 11, the storage 13, or another recording medium. For example, a tape, a disk, a card, a semiconductor memory, or a programmable logic circuit may be used as the other recording medium.

(Testing method flow)
The flow of the inspection method S1 performed by the inspection device 1 will be described with reference to Fig. 2. Fig. 2 is a flow chart showing the flow of the inspection method S1.

As shown in FIG. 2, the inspection method S1 includes a specification process S11, a generation process S12, and an inspection process S13. In the following description, it is assumed that an input image Img containing multiple objects O1, O2, ..., On as subjects is stored in the memory 11 of the inspection device 1.

The identification process S11 is a process for identifying multiple attention areas A1, A2, ..., An in the input image Img, each of which contains a single object Oi (i = 1, 2, ..., n). The attention area Ai is an area in the input image Img that contains the corresponding object Oi, and is preferably set so that (1) the image of the corresponding object Oi does not have any common parts with the outside of the attention area Ai, and (2) the images of other objects Oj (j ≠ i) do not have any common parts with the inside of the attention area Ai. In this embodiment, the identification process S11 is executed by the processor 12 provided in the inspection device 1.

The generation process S12 is a process for generating an image Imgi including each attention area Ai identified in the identification process S11. In this embodiment, the generation process S12 is executed by the processor 12 provided in the inspection device 1.

The inspection process S13 is a process for performing image inspection of an object Oi contained as a subject in each image Imgi generated in the generation process S12, based on that image Imgi. In this embodiment, the inspection process S13 is executed by the processor 12 provided in the inspection device 1 using the trained model M.

The trained model M is a model that takes an image containing an object as a subject as input and outputs the class to which the object belongs. The trained model M is constructed, for example, by machine learning using pairs of images containing an object (an object whose class is known) as a subject and the class to which the object belongs as training data.

(Specific examples of specific processing)
A specific example of the specific process S11 included in the inspection method S1 will be described with reference to Fig. 3. Fig. 3 is a flow chart showing the flow of the specific process S11 according to this specific example.

Depending on the state of the object Oi or the state of the image of the object Oi in the input image Img, it may be difficult to identify the attention area Ai containing the object Oi using template matching. The identification process S11 according to this specific example is devised so that even in such a case, it is possible to identify the attention area Ai containing the object Oi for all objects O1, O2, ..., On included as subjects in the input image Img. For the sake of convenience, n-m objects O1, O2, ..., On-m are defined as objects whose attention area can be identified by template matching, and m objects On-m+1, On-m+2, ..., On are defined as objects whose attention area cannot be identified by template matching. Here, m is a natural number between 1 and n-2, and represents the number of objects whose attention area cannot be identified by template matching.

As shown in FIG. 3, the specific process S11 in this specific example includes a first specific process S111 and a second specific process S112.

The first identification process S111 is a process for identifying nm-m regions of interest A1, A2, ..., An-m, each of which contains a single object Oj (j = 1, 2, ..., nm) in the input image Img, using template matching. Each region of interest Aj identified in the first identification process S111 is an area in the input image Img that contains the object Oj. In this embodiment, the first identification process S111 is executed by the processor 12 provided in the inspection device 1.

The second identification process S112 is a process for identifying m areas of interest An-m+1, An-m+2, ..., An in the input image Img that are different from the areas of interest A1, A2, ..., An-m identified in the first identification process S111, using the periodicity of the n-m areas of interest A1, A2, ..., An-m identified in the first identification process S111. When objects O1, O2, ..., On are periodically arranged in the input image Img, each area of interest Ak identified in the second identification process S112 is an area that includes the corresponding object Ok in the input image Img, similar to the area of interest Aj identified in the first identification process S111. In this embodiment, the second identification process S112 is executed by the processor 12 provided in the inspection device 1.

The second identification process S112 can be composed of a first calculation process S112a and a second calculation process S112b, as shown in FIG. 3.

The first calculation process S112a is a process for calculating the distance dx between the representative points of two adjacent attention areas aligned in the pixel row direction (an example of the "first direction" in the claims) among the nm-m attention areas A1, A2, ..., An-m identified in the first identification process S111, and the distance dy between the representative points of two adjacent attention areas aligned in the pixel column direction (an example of the "second direction" in the claims) among the nm-m attention areas A1, A2, ..., An-m identified in the first identification process S111.

The second calculation process S112b is a process for calculating the coordinates of representative points of m attention areas An-m+1, An-m+2, ..., An that are different from the n-m attention areas A1, A2, ..., An-m identified in the first identification process S111, based on the intervals dx, dy calculated in the first calculation process S112a. The m attention areas An-m+1, An-m+2, ..., An are set as attention areas whose coordinates of representative points match the coordinates calculated in the second calculation process S112b.

Figure 4 is a diagram showing an example of execution of the identification process S11 according to this specific example. Here, as shown in (a) of Figure 4, consider an input image Img that includes four objects O1, O2, O3, and O4 as subjects. The three objects O1, O2, and O3 are objects for which it is possible to identify the region of interest by template matching, while the remaining object O4 is an object for which it is impossible to identify the region of interest by template matching due to the presence of a foreign matter or the like.

In the first identification process S111, as shown in FIG. 4B, three attention areas A1, A2, and A3 that respectively include three objects O1, O2, and O3 are identified by template matching.

In the first calculation process S112a of the second identification process S112, as shown in FIG. 4(c), the distance dx between the representative points P1, P2 of two adjacent attention areas A1, A2 aligned in the pixel row direction, and the distance dy between the representative points P1, P3 of two adjacent attention areas A1, A3 aligned in the pixel column direction are calculated. Here, each attention area Ai is defined as a rectangular area having a side parallel to the pixel row direction and a side parallel to the pixel column direction, and the representative point Pi of each attention area Ai is defined as the vertex of the upper left corner of the rectangular area.

In the second calculation process S112b of the second identification process S112, as shown in FIG. 4D, the coordinates of the representative point P4 of the remaining attention area A4 are calculated based on the intervals dx and dy calculated in the first calculation process S112a. Specifically, the x coordinate (coordinate in the pixel row direction) of the representative point P4 of the attention area A4 is calculated by adding the interval dx to the coordinate in the pixel row direction of the representative point P3 of the attention area A3, and the y coordinate (coordinate in the pixel column direction) of the representative point P4 of the attention area A4 is calculated by adding the interval dy to the coordinate in the pixel column direction of the representative point P2 of the attention area A2. Then, the attention area A4 is set to a rectangular area whose coordinates of the vertex of the upper left corner match the coordinates calculated in the second calculation process S112b.

(Specific example of generation process)
A specific example of the generation process S12 included in the inspection method S1 will be described with reference to Fig. 5. Fig. 5 is a flow diagram showing the flow of the generation process S12 according to this specific example.

As shown in FIG. 5, the generation process S12 in this specific example includes a cut-out process S121 and an embedding process S122.

The cut-out process S121 is a process for cutting out each attention area Ai identified in the identification process S11 from the input image Img. The embedding process S122 is a process for embedding each attention area Ai cut out in the cut-out process S121 in an image Imgi of a predetermined size without processing. Here, "embedding without processing" the attention area Ai refers to embedding the attention area Ai cut out from the input image Img as it is (without changing the pixel values of all pixels) without performing image processing such as enlargement, reduction, and rotation. As a result, an image Imgi of a predetermined size is generated for each attention area Ai, in which the attention area Ai is embedded without processing. In the image Imgi generated in the cut-out process S121, pixels other than the part where the attention area Ai is embedded are set to the same pixel value. The pixel values of these pixels are arbitrary, and may be, for example, a pixel value corresponding to black or a pixel value corresponding to white.

Figure 6 shows a specific example of an image Imgi generated by generation process S12 in this specific example.

Image Img1 shown in FIG. 6(a) is an image of 600 pixels by 1200 pixels. Area of interest A1 is a rectangular area in the input image Img with sides parallel to the pixel row direction and sides parallel to the pixel column direction. This area of interest A1 is embedded as is in image Img1 without being enlarged, reduced, or rotated. In image Img1, the pixel values of pixels other than the part where area of interest A1 is embedded are uniformly set to a pixel value that represents black.

Image Img2 shown in FIG. 6(b) is an image of 600 pixels by 1200 pixels, similar to image Img1 shown in FIG. 6(a). Area of interest A2 is a rectangular area in the input image Img with sides parallel to the pixel row direction and sides parallel to the pixel column direction. This area of interest A2 is smaller than area of interest A1, but this area of interest A2 is embedded in image Img2 as is, without being enlarged, reduced, or rotated. In image Img2, the pixel values of pixels other than the part where area of interest A2 is embedded are set uniformly to pixel values representing black, similar to image Img1.

Image Img3 shown in FIG. 6(c) is an image of 600 pixels by 1200 pixels, similar to image Img1 shown in FIG. 6(a) and image Img2 shown in FIG. 6(b). Area of interest A3 is a rectangular area in the input image Img that has sides that are non-parallel to the pixel row direction and sides that are non-parallel to the pixel column direction. This area of interest A3 is tilted, but it is embedded as is in image Img3 without being enlarged, reduced, or rotated. In image Img3, the pixel values of pixels other than the part where area of interest A3 is embedded are set uniformly to pixel values that represent black, similar to images Img1 and Img2.

(Examples of application)
The inspection apparatus 1 according to this embodiment can be applied to, for example, the inspection of a plurality of laser diode chips formed on a single wafer. An input image obtained by imaging such a wafer is shown in FIG. 7A.

The template used for template matching may be a template that matches the entire laser diode, or a template that matches a portion of the laser diode. Examples of the latter template include a template that matches the top and bottom ends of the laser diode, or a template that matches the right and left ends of the laser diode. The area detected by template matching using a template that matches the top and bottom ends of the laser diode is shown in FIG. 7(b). In FIG. 7(b), the area detected by template matching is shown surrounded by a dotted rectangle.

When a template that matches the entire laser diode is used, the area detected by template matching can be identified as the area of interest. On the other hand, when a template that matches part of the laser diode is used, the area derived from the area detected by template matching according to a specified algorithm can be identified as the area of interest. For example, when a template that matches the upper and lower ends of the laser diode is used, a rectangular area that spans two adjacent areas lined up in the pixel column direction among the areas detected by template matching can be identified as the area of interest. The area of interest identified in this way is shown in FIG. 7(c). In FIG. 7(c), the area of interest identified in this way is shown surrounded by a solid-line rectangle.

Figure 7(d) shows an input image obtained by imaging a wafer with foreign matter attached. In such a case, template matching may fail for the area where the foreign matter is attached. However, as described above, by identifying the area of interest in two stages by the first identification process S111 and the second identification process S112, the possibility of failing to set the area of interest in the area where the foreign matter is attached can be significantly reduced.

(Effect 1 of the inspection device)
As described above, the inspection device 1 according to the present embodiment includes a processor 12 that executes the first identification process S111, the second identification process S112, and the inspection process S13. The first identification process S111 is a process for identifying nm-m attention regions A1, A2, ..., An-m, each of which includes a single object Oj (j=1, 2, ..., nm) in an input image Img including n objects O1, O2, ..., On as subjects. The second identification process S112 is a process for identifying m attention regions An-m+1, An-m+2, ..., An that are different from the nm-m attention regions A1, A2, ..., An-m identified in the first identification process S111, based on the periodicity of the nm-m attention regions A1, A2, ..., An-m identified in the first identification process S111. The inspection process S13 is a process for performing image inspection of an object Oi contained in an area of interest Ai (i = 1, 2, ..., n) based on each of the nm areas of interest A1, A2, ..., An-m identified in the first identification process S111 and each of the m areas of interest An-m+1, An-m+2, ..., An identified in the second identification process S112.

According to the above configuration, when objects O1, O2, ..., On are arranged periodically in the input image Img, even if the first identification process S111 fails to identify the attention areas An-m+1, An-m+2, ..., An that include the objects On-m+1, On-m+2, ..., On, the attention areas An-m+1, An-m+2, ..., An that include these objects On-m+1, On-m+2, ..., On can be identified in the second identification process S112. Therefore, it is possible to realize an inspection device 1 that can inspect all objects O1, O2, ..., On included as subjects in the input image Img without omission. The same effect can be obtained by the above-mentioned inspection method S1. The same effect can also be obtained by executing the above-mentioned inspection program P on a computer.

In addition, in the inspection device 1 according to this embodiment, the processor 12 executes the first identification process S111 using template matching.

The above configuration makes it possible to realize an inspection device 1 that can efficiently identify areas of interest A1, A2, ..., An-m that contain objects O1, O2, ..., On-m in the first identification process S111.

In addition, in the inspection device 1 according to this embodiment, the processor 12 executes the inspection process S13 using a trained model M that receives as input the nm regions of interest A1, A2, ..., An-m identified in the first identification process S111 and the m regions of interest An-m+1, An-m+2, ..., An identified in the second identification process S112, and outputs the class to which the object Oi contained in the input region of interest Ai belongs.

The above configuration makes it possible to realize an inspection device 1 that can accurately inspect each of the objects O1, O2, ..., On contained as subjects in the input image Img.

(Effect of the inspection device 2)
As described above, the inspection device 1 according to the present embodiment includes a processor 12 that executes a specification process S11, a generation process S12, and an inspection process S13. The specification process S11 is a process of specifying n attention areas A1, A2, ..., An, each of which includes a single object Oj, in an input image Img including n objects O1, O2, ..., On as subjects. The generation process S12 is a process of generating an image Imgi (i = 1, 2, ..., n) of a predetermined size for each of the n attention areas A1, A2, ..., An specified in the specification process S11, in which the attention area Ai is embedded without processing. The inspection process S13 is a process of performing an image inspection of an object Oi included in the attention area Ai embedded in the image Imgi, based on each of the n images Img1, Img2, ..., Imgn generated in the generation process S12.

According to the above configuration, the images Img1, Img2, ..., Imgn referenced in the inspection process S13 have a uniform size. Furthermore, according to the above configuration, an area of interest Ai that has not been degraded by processing is embedded in each image Imgi referenced in the inspection process S13. Therefore, it is possible to realize an inspection device 1 that can execute the inspection process S13 with high accuracy according to a specific algorithm. The same effect can be obtained by the above-mentioned inspection method S1. Furthermore, the same effect can be obtained by having a computer execute the above-mentioned inspection program P.

In addition, in the inspection device 1 according to this embodiment, the processor 12 executes the first identification process S111 using template matching.

The above configuration makes it possible to realize an inspection device 1 that can efficiently identify areas of interest A1, A2, ..., An that contain objects O1, O2, ..., On in the first identification process S111.

In addition, in the inspection device 1 according to this embodiment, in the generation process S12, the processor 12 generates an image Imgi in which the pixel values of pixels other than the portion in which the attention area Ai is embedded are the same.

The above configuration can reduce the influence on the inspection results of the pixel values of pixels other than the part in which the attention area Ai is embedded in the image Imgi. Therefore, it is possible to realize an inspection device 1 that can accurately inspect each of the objects O1, O2, ..., On included as subjects in the input image Img.

In addition, in the inspection device 1 according to this embodiment, the processor 12 executes the inspection process S13 using a trained model M that receives as input each of the images Img1, Img2, ..., Imgn generated in the generation process S12, and outputs the class to which the object Oi contained in the attention area Ai embedded in the input image Imgi belongs.

The above configuration makes it possible to realize an inspection device 1 that can accurately inspect each of the objects O1, O2, ..., On contained as subjects in the input image Img.

(Additional Notes)
The present invention is not limited to the above-described embodiments, and various modifications are possible within the scope of the claims. Embodiments obtained by appropriately combining the technical means disclosed in the above-described embodiments are also included in the technical scope of the present invention.

1 Inspection device 11 Memory 12 Processor 13 Storage P Inspection program M Trained model

Claims (6)

a first identification process for identifying a plurality of interest regions each including a single object in an input image including a plurality of objects as subjects;
a second identification process for identifying, in the input image, a single or multiple attention regions each including a single object, different from the attention regions identified in the first identification process, based on a periodicity of the multiple attention regions identified in the first identification process;
and an inspection process for performing an image inspection of an object included in each of the plurality of attention areas identified in the first identification process and the single or plurality of attention areas identified in the second identification process.
An inspection device characterized by: The processor executes the first identification process using template matching.
2. The inspection apparatus according to claim 1 . In the second identification process, the processor executes: (1) a first calculation process to calculate one or both of an interval between representative points of two adjacent attention areas aligned in a first direction among the multiple attention areas identified in the first identification process, and an interval between representative points of two adjacent attention areas aligned in a second direction intersecting the first direction among the multiple attention areas identified in the first identification process; and (2) a second calculation process to calculate, based on the interval calculated in the first calculation process, coordinates of a representative point of a attention area other than the attention area identified in the first identification process.
3. The inspection apparatus according to claim 1 or 2. The processor executes the inspection process using a trained model that receives as input each of the multiple regions of interest identified in the first identification process and the single or multiple regions of interest identified in the second identification process, and outputs a class to which an object included in the input region of interest belongs.
4. The inspection device according to claim 1, wherein the inspection device is a semiconductor laser. A first identification process in which a single or multiple processors identify a plurality of interest regions each including a single object in an input image including a plurality of objects as subjects;
a second identification process in which the processor identifies, in the input image, a single or multiple attention regions, each of which includes a single object, different from the attention regions identified in the first identification process, based on a periodicity of the multiple attention regions identified in the first identification process;
and an inspection process in which the processor performs an image inspection of an object included in each of the plurality of attention regions identified in the first identification process and the single or plurality of attention regions identified in the second identification process.
13. An inspection method comprising: An inspection program for causing a computer to operate as the inspection device according to any one of claims 1 to 4, the inspection program causing a processor included in the computer to execute each of the processes.

Images