JP2020102111A - Information processing device and visual inspection device - Google Patents

Abstract

To improve inspection accuracy when visually inspecting an inspection object using a learning model.SOLUTION: A symmetrical part image is extracted from an image for learning that is obtained by imaging the same type of an object as an inspection object, and the symmetrical part image is halved in a symmetrical direction to generate divided images, the divided images being further inverted to generate inverted images. One of the divided images and an inverted image from the other of the divided images obtained by inversion are acquired as teaching data for one side of the symmetrical part of the inspection object. Also, the other of the divided images and an inverted image from the one of the divided images obtained by inversion are acquired as teaching data for the other side of the symmetrical part of the inspection object. Then, machine learning is executed on the basis of the acquired teaching data, and a learning model used for the visual inspection of the inspection object is generated thereby.SELECTED DRAWING: Figure 3

Description

The present invention relates to an information processing device and a visual inspection device.
Regarding

Conventionally, various techniques have been developed for an inspection apparatus that inspects the appearance of the inspection target using an image of the inspection target. For example, Patent Literature 1 discloses a technique for performing an inspection in an inspection apparatus by dividing a captured image of an inspection target into a plurality of captured divided images. In the technique described in Patent Document 1, a feature amount is extracted by performing machine learning on each of a plurality of learning divided images obtained by dividing the learning image into a plurality of images. Further, each of the plurality of captured divided images is classified based on the extracted feature amount. Then, the quality of the inspection object is determined based on the plurality of classified imaged divided images.

JP, 2017-212159, A

It is conceivable to inspect the appearance of the inspection target by using a learning model constructed by performing machine learning using a learning image obtained by capturing an object of the same type as the inspection target as teacher data. At this time, in order to improve the inspection accuracy, it is necessary to construct a learning model based on more teacher data.

The present invention has been made in view of the above problems, and an object of the present invention is to provide a technique capable of improving inspection accuracy when performing a visual inspection of an inspection target using a learning model. And

The information processing apparatus according to the first aspect of the present invention is used for appearance inspection of an inspection target object by performing machine learning using learning images obtained by capturing an object of the same type as the inspection target object as teacher data. In an information processing device for generating a learning model, an image of a symmetrical portion which is a symmetrical portion or a vertically symmetrical portion of the object is extracted from the learning image, and the image of the symmetrical portion is divided into two in the symmetrical direction. A divided image is thereby generated, and further, a first image processing unit that generates a reversed image by reversing the image on one side and the image on the other side in the divided image, and an image on one side in the divided image, And, the inverted image obtained by inverting the image on the other side in the divided image is acquired as the teacher data for one side in the symmetrical portion, and the image on the other side in the divided image, and in the divided image. A learning unit that acquires the inverted image obtained by inverting the image on one side as the teacher data for the other side in the symmetrical portion, and performs the machine learning based on the acquired teacher data to generate the learning model. And

The information processing apparatus according to the present invention performs learning by machine learning using a learning image obtained by imaging an object of the same type as the inspection object to generate a learning model used for appearance inspection of the inspection object. .. In this case, the accuracy of the learning model can be improved by acquiring more learning images that can serve as teacher data.

Then, in the information processing device according to the present invention, the first image processing unit extracts an image of a symmetrical portion of the imaged object from the learning image for generating the learning model. Here, the symmetric portion of the imaged object is a symmetrical portion or a vertically symmetrical portion of the inspection object. The symmetrical portion of the inspection object can be extracted based on the design value of the inspection object or the like.

Further, the first image processing unit acquires a divided image by dividing the extracted image of the symmetrical portion into two in the symmetrical direction. That is, when the symmetrical portion is a left-right symmetrical portion, the image of the symmetrical portion is divided into the image on the right side and the image on the left side. When the symmetrical portion is a vertically symmetrical portion, the image of the symmetrical portion is divided into the upper portion image and the left portion image. Furthermore, the first image processing unit inverts the image on one side and the image on the other side in the divided image to generate an inverted image.

Then, the learning unit acquires an image on one side of the divided image and an inverted image obtained by inverting the image on the other side of the divided image as teacher data for one side of the symmetrical portion. That is, one divided image that is one side image when the image of the symmetrical portion of the captured object is divided into two, and the other side image when the image of the symmetrical portion of the captured object is divided into two. An inverted image obtained by inverting the divided image on the side is acquired as teacher data for extracting the feature amount used for the visual inspection of the one side of the symmetrical portion.

Further, the learning unit acquires the image on the other side of the divided image and the inverted image obtained by inverting the image on the one side of the divided image as the teacher data for the other side of the symmetrical portion. That is, one of the divided image on the other side, which is the image on the other side when the image of the symmetrical portion of the captured object is divided into two, and the image on the one side, when the image of the symmetrical portion of the captured object is divided into two. An inverted image obtained by inverting the divided image on the side is acquired as teacher data for extracting the feature amount used for the visual inspection on the other side of the symmetrical portion.

Here, the appearances of one part and the other part of the symmetrical portion of the inspection object are, of course, the same in designing the inspection object. Therefore, in designing an object of the same type as the inspection object (imaged object), when one of the image of the symmetrical portion of the object and the image of the other portion are inverted and overlapped, , Both will be the same. Therefore, an inverted image obtained by inverting the image on one side in the divided image can be used as teacher data for the other side in the symmetrical portion of the inspection object, and the inverted image obtained by inverting the image on the other side in the divided image can be used. The image can be used as teacher data for one side of the symmetrical portion.

Then, as described above, it is possible to increase the teacher data for constructing the learning model by adding the inverted images of the one side image and the other side image in the divided image as the teacher data. As a result, the accuracy of the learning model can be improved, and thus the inspection accuracy when performing the visual inspection of the inspection object using the learning model can be improved.

An appearance inspection apparatus according to a second aspect of the present invention is an appearance inspection apparatus that performs an appearance inspection of an inspection object using the learning model generated by the information processing apparatus according to the first aspect. From the inspection object image obtained by imaging the inspection object, an image of a symmetrical portion which is a symmetrical portion or a vertically symmetrical portion of the inspection object is extracted, and the symmetry extracted from the inspection object image A second image processing unit that generates a divided image by dividing the partial image into two in the symmetric direction, the divided image generated from the inspection target image is acquired as inspection data, and the acquired inspection data is acquired. An inspection unit that determines whether the appearance of the symmetrical portion of the inspection object is acceptable by inputting the learning model.

The information processing method according to the third aspect of the present invention uses the learning image obtained by imaging an object of the same type as the inspection target as machine data to perform machine learning, thereby performing an appearance inspection of the inspection target. In the information processing method for generating the learning model used for, from the learning image, a step of extracting an image of a symmetrical portion which is a symmetrical portion or a vertically symmetrical portion of the object, and the image of the symmetrical portion in the symmetrical direction. A step of generating a divided image by dividing the image into two, a step of generating an inverted image by inverting each of the image on one side and the image on the other side in the divided image, an image on one side in the divided image, and , The inverted image obtained by inverting the image on the other side in the divided image is acquired as the teacher data for one side in the symmetrical portion, and the image on the other side in the divided image, and one in the divided image Acquiring the inverted image obtained by inverting the image on the side as the teacher data for the other side in the symmetrical portion, and generating the learning model by performing machine learning based on the acquired teacher data. And.

According to the present invention, it is possible to improve the inspection accuracy when performing the visual inspection of the inspection target using the learning model.

It is a block diagram which shows the outline of a function structure of the vehicle inspection device which concerns on embodiment. It is a figure for explaining an example of image processing performed in an image processing part when performing machine learning. It is a flow chart which shows a flow of processing when machine learning is performed in a vehicle inspection device. It is a flow chart which shows a flow of processing at the time of performing a visual inspection of a vehicle under inspection in a vehicle inspection device.

Hereinafter, specific embodiments of the present invention will be described with reference to the drawings. The dimensions, materials, shapes, relative arrangements, and the like of the components described in this embodiment are not intended to limit the technical scope of the invention to these unless otherwise specified.

<Embodiment>
(Structure of vehicle inspection device)
In the present embodiment, the present invention is applied to a vehicle inspection device that inspects the appearance of a vehicle. That is, in this embodiment, the “vehicle” corresponds to the “inspection object”. Here, the functional configuration of the vehicle inspection device according to the present embodiment will be described based on FIG. 1. FIG. 1 is a block diagram showing an outline of a functional configuration of a vehicle inspection device according to this embodiment. The vehicle inspection device 100 is a device that performs an appearance inspection of the vehicle on the basis of an image of the vehicle to be inspected (inspection target image) captured by the image capturing device 200. In addition, the vehicle inspection device 100 performs machine learning based on a learning image obtained by capturing an image of a vehicle of the same type as the vehicle to be inspected, thereby generating a learning model to be used for the appearance inspection of the vehicle to be inspected. Have the function to

The vehicle inspection device 100 includes a processor such as a CPU and a DSP, a main storage unit such as a read-only memory (ROM) and a random access memory (RAM), and an auxiliary storage unit such as an EPROM, a hard disk drive (HDD), and a removable medium. , And a computer having. Here, the removable medium is, for example, a flash memory such as a USB memory or an SD card, or a disc recording medium such as a CD-ROM, a DVD disc, or a Blu-ray disc.

Further, in the vehicle inspection device 100, the processor loads the program stored in the auxiliary storage unit into the main storage unit and executes the program, so that various functions described below can be realized. However, some or all of the functions of the vehicle inspection device 100 may be implemented by a hardware circuit such as an ASIC or FPGA. Further, as the imaging device 200, a wearable camera or the like worn by an operator who inspects the vehicle can be used.

The vehicle inspection device 100 and the imaging device 200 are connected to each other via a network. For this network, for example, a WAN (Wide Area Network), which is a world-wide public communication network such as the Internet, or a telephone communication network such as a mobile phone may be adopted. In addition, a network may be constructed by using a communication standard such as Bluetooth (registered trademark) Low Energy standard or WiFi (registered trademark).

The vehicle inspection device 100 includes an image processing unit 110, a learning unit 120, and an inspection unit 130. The image processing unit 110 has a function of performing image processing necessary for a visual inspection on an inspection target image captured by the image capturing apparatus 200. The image processing unit 110 also has a function of performing image processing necessary for machine learning on the learning image captured by the imaging device 200. The inspection target image and the learning image are, for example, a JPEG (Joint Photographic Experts Group), a bitmap image (bitmap image), a GIF (Graphics Interchange Format) data format table, or the like.

The image processing unit 110 includes an image acquisition unit 111, a symmetric portion extraction unit 112, a divided image generation unit 113, and a reverse image generation unit 114 as functional units that execute various processes in image processing. The details of the processing executed by these functional units will be described later.

The learning unit 120 has a function of generating a learning model by performing machine learning using the learning image that has undergone the necessary image processing in the image processing unit 110. The learning unit 120 has a teacher data acquisition unit 121 and a learning model generation unit 122 as functional units that execute various processes in machine learning. In addition, the inspection unit 130 inputs the inspection target image, which has undergone the necessary image processing in the image processing unit 110, into the learning model generated in the learning unit 120 to perform the appearance inspection of the inspection target vehicle. Have. The inspection unit 130 includes a learning model storage unit 131, an inspection data acquisition unit 132, and a determination unit 133 as functional units that execute various processes in the appearance inspection. The details of the processing executed by each functional unit of the learning unit 120 and the inspection unit 130 will also be described later.

(Machine learning)
Here, in the vehicle inspection device 100, a process executed by the image processing unit 110 and the learning unit 120 when the machine learning is performed to generate the learning model used for the visual inspection of the vehicle will be described. First, image processing executed in the image processing unit 110 when performing machine learning will be described based on FIG. 2 in addition to FIG. FIG. 2 is a diagram for explaining an example of image processing executed in the image processing unit 110 when performing machine learning. FIG. 2 shows a flow of image processing when performing machine learning for generating a learning model used for appearance inspection of a vehicle body rear surface including a back door of a vehicle.

When machine learning is performed, a learning image is captured by capturing an image of a vehicle of the same type as the vehicle to be inspected by the image capturing device 200. Then, the image acquisition unit 111 in the image processing unit 110 acquires the learning image captured by the imaging device 200. Figure 2(a) shows
The learning image Im1 (image on the rear surface of the vehicle) acquired by the image acquisition unit 111 is illustrated.

The symmetric portion extraction unit 112 of the image processing unit 110 extracts an image of a symmetric portion of the vehicle from the learning image acquired by the image acquisition unit 111. Here, the symmetrical portion is a portion that is symmetrical in the vehicle or vertically symmetrical in the vehicle. FIG. 2B shows an image Im2 (image on the rear surface of the vehicle body) of the symmetrical portion extracted by the symmetrical portion extraction unit 112 from the learning image Im1 shown in FIG. 2A. Here, in terms of vehicle design, the back surface of the vehicle body is symmetrical. Therefore, the symmetrical portion extraction unit 112 extracts the image Im2 on the back surface of the vehicle body as an image of a symmetrical portion of the vehicle. Note that design information regarding design values of the vehicle to be inspected may be stored in the vehicle inspection device 100 in advance. Then, the symmetrical portion extraction unit 112 may select a symmetrical portion of the vehicle based on the stored design information and extract the image of the selected symmetrical portion from the learning image.

The divided image generation unit 113 in the image processing unit 110 divides the image of the symmetrical portion of the vehicle extracted by the symmetrical portion extraction unit 112 into two in the symmetrical direction to generate a divided image. FIG. 2C shows divided images Im3 and Im4 obtained by dividing the image Im2 on the rear surface of the vehicle body, which is the image of the symmetrical portion shown in FIG. 2B, into a right side portion and a left side portion. As described above, when the symmetrical portion is a left-right symmetrical portion, the divided image generation unit 113 divides the image of the symmetrical portion into the image of the right side portion and the image of the left side portion. If the symmetrical portion is a vertically symmetrical portion, the divided image generation unit 113 divides the image of the symmetrical portion into an image of the upper portion and an image of the left portion.

The reverse image generation unit 114 in the image processing unit 110 generates a reverse image by inverting each of the divided images generated by the divided image generation unit 113. 2D shows an inverted image Im5 obtained by inverting the divided image Im3 on the right side of the symmetrical portion shown in FIG. 2C, and an divided image Im4 on the right side of the symmetrical portion shown in FIG. 2C. The reversed image Im6 which was reversed is shown. In this way, when the divided image is an image obtained by dividing the image of the symmetrical portion into the image of the right side portion and the image of the left side portion, the reverse image generation unit 114 reverses each divided image in the horizontal direction. On the other hand, when the divided image is an image obtained by dividing the image of the symmetrical portion into the image of the upper portion and the image of the lower portion, the reverse image generation unit 114 reverses each divided image in the vertical direction.

In the following description, in the divided images generated by the divided image generating unit 113, one side image may be referred to as “first divided image” and the other side image may be referred to as “second divided image”. is there. Further, in the reverse image generated by the reverse image generating unit 114, the image obtained by inverting the first divided image is referred to as “first inverted image”, and the image obtained by inverting the second divided image is referred to as “first image”. It may also be referred to as "two-inverted image". For example, when the divided image Im3 in FIG. 2C is the first divided image, the reverse image Im5 in FIG. 2D corresponds to the first reverse image. When the divided image Im4 in FIG. 2C is the second divided image, the reverse image Im6 in FIG. 2D corresponds to the second reverse image.

Next, a learning model generation process executed by the learning unit 120 when performing machine learning will be described. The teacher data acquisition unit 121 in the learning unit 120 acquires the divided image and the inverted image generated by the image processing in the image processing unit 110 as teacher data for machine learning. More specifically, the teacher data acquisition unit 121 uses the first divided image generated by the divided image generation unit 113 and the second inverted image generated by the inverted image generation unit 114 as one side in the symmetrical portion of the vehicle. Is acquired as teacher data. That is, the teacher data acquisition unit 121 acquires the first divided image and the second inverted image as teacher data for extracting the feature amount used for the visual inspection of the one side of the symmetrical portion of the vehicle. Further, the teacher data acquisition unit 121 uses the second divided image generated by the divided image generation unit 113,
Also, the first reverse image generated by the reverse image generating unit 114 is acquired as teacher data for the other side of the symmetrical portion of the vehicle. That is, the teacher data acquisition unit 121 acquires the second divided image and the first inverted image as teacher data for extracting the feature amount used for the visual inspection on the other side of the symmetrical portion of the vehicle.

Therefore, for example, when the image processing apparatus 100 performs image processing as shown in FIG. 2, the teacher data acquisition unit 121 determines the divided image Im3 that is the first divided image and the inverted image Im6 that is the second inverted image. , The divided image Im4 which is the second divided image and the inverted image Im5 which is the first inverted image, are acquired as teacher data for extracting the feature amount used for the appearance inspection of the right side portion of the vehicle body rear surface. It is acquired as teacher data for extracting the feature amount used for the appearance inspection of the.

In the vehicle shown in FIG. 2, the appearances of the right side portion and the left side portion of the vehicle body rear surface, which are symmetrical portions, are the same in design. That is, in terms of vehicle design, when an image of the right side portion of the rear surface of the vehicle body and an image obtained by inverting the image of the left side portion of the vehicle body are overlapped, the two are in agreement. Further, in terms of vehicle design, when an image of the left side portion of the back of the vehicle body and an image obtained by inverting the image of the right side portion of the vehicle body are overlapped, the two are in agreement. Therefore, not only the divided image Im3 obtained by capturing the right side portion of the vehicle body rear surface, but also the inverted image Im6 of the divided image Im4 obtained by capturing the left side portion of the vehicle body can be used as teacher data for the right side portion of the vehicle body rear surface. Further, not only the divided image Im4 obtained by capturing the left side portion of the vehicle body rear surface but also the inverted image Im5 of the divided image Im3 obtained by capturing the right side portion of the vehicle body rear surface can be used as teacher data for the left side portion of the vehicle body rear surface.

Then, the learning model generation unit 122 in the learning unit 120 generates a learning model used for the visual inspection of the vehicle based on the teacher data acquired by the teacher data acquisition unit 121. That is, for the symmetrical portion of the vehicle, the learning model is generated using the divided image and the inverted image as teacher data.

According to the above, in machine learning for constructing the learning model used for the visual inspection of the vehicle, not only the divided images obtained by dividing the imaged image of the symmetrical portion of the vehicle but also the image on one side and the image on the other side of the divided image. By using the inverted image of each image as the teacher data, it is possible to increase the teacher data for constructing the learning model. As a result, the accuracy of the learning model can be improved.

Hereinafter, a flow of processing when machine learning is performed in the vehicle inspection device 100 will be described based on the flowchart shown in FIG. The processing of S101 to S104 in this flow is executed by the image processing unit 110, and the processing of S105 and S106 is executed by the learning unit 120.

In this flow, in S101, the image acquisition unit 111 acquires the image of the vehicle captured by the image capturing apparatus 200 as a learning image. Next, in S102, the symmetrical portion extraction unit 112 extracts an image of a symmetrical portion of the vehicle from the learning image acquired by the image acquisition unit 111. Next, in S103, the divided image generation unit 113 generates a divided image by dividing the image extracted by the symmetric portion extraction unit 112 into two in the symmetric direction. Next, in S104, the reverse image generation unit 114 generates a reverse image by inverting each reverse image generated by the divided image generation unit 113.

Next, in S105, the teacher data acquisition unit 121 acquires, as the teacher data, the divided image generated by the divided image generation unit 113 and the inverted image generated by the inverted image generation unit 114. Next, in S106, the learning model generation unit 122 generates a learning model used for the visual inspection of the vehicle based on each teacher data (that is, the divided image and the inverted image) acquired by the teacher data acquisition unit 121.

Then, when a learning model is generated by the learning model generation unit 122 by executing the above flow, the learning model is stored in the learning model storage unit 131 of the inspection unit 130.

(Visual inspection)
Next, in the vehicle inspection device 100, a process executed by the image processing unit 110 and the inspection unit 130 when the appearance inspection of the inspection target vehicle is performed will be described. When the visual inspection of the vehicle to be inspected is performed, the image of the vehicle to be inspected is captured by the image capturing device 200, and thus the image to be inspected is captured. Then, the image acquisition unit 111 in the image processing unit 110 acquires the inspection target image captured by the imaging device 200.

When the image acquisition unit 111 acquires the inspection target image, the symmetrical portion extraction unit 112 extracts an image of a symmetrical portion of the vehicle from the inspection target image. Then, when the image of the symmetrical portion of the vehicle is extracted from the inspection target image by the symmetrical portion extracting unit 112, the divided image generating unit 113 divides the image of the symmetrical portion into two in the symmetrical direction, and thus the divided image is obtained. To generate. As described above, when the visual inspection of the vehicle to be inspected is performed, the image acquisition unit 111, the symmetric portion extraction unit 112, and the divided image generation unit 113 of the image processing unit 110 perform the inspection target images. The processing is similar to the processing executed on the learning image by each functional unit of the image processing unit 110 when the above-described machine learning is performed.

Then, when the image processing unit 110 performs the above-described image processing on the inspection target image, the inspection data acquisition unit 132 in the inspection unit 130 sets the divided image generated by the image processing to the inspection target image. Acquired as inspection data for vehicle visual inspection. When the inspection data acquisition unit 132 acquires the inspection data, the determination unit 133 inputs the inspection data into the learning model stored in the learning model storage unit 131, so that the symmetrical portion of the vehicle to be inspected Pass/Fail is determined regarding the appearance. That is, the pass/fail is determined for the appearances on one side and the other side of the symmetrical portion of the vehicle to be inspected.

According to this, the visual inspection of the vehicle to be inspected can be performed by the learning model constructed by using the divided image and the inverted image generated based on the learning image as the teacher data. That is, it is possible to perform a visual inspection of the vehicle using a learning model with higher accuracy. Therefore, the inspection accuracy of the visual inspection of the vehicle can be improved.

Hereinafter, a flow of processing when the vehicle inspection device 100 performs an appearance inspection of a vehicle to be inspected will be described based on a flowchart shown in FIG. The processing of S201 to S203 in this flow is executed by the image processing unit 110, and the processing of S204 and S205 is executed by the inspection unit 130.

In this flow, in S201, the image acquisition unit 111 acquires the image of the vehicle as the inspection target captured by the image capturing apparatus 200 as the inspection target image. Next, in S202, the symmetrical portion extraction unit 112 extracts an image of a symmetrical portion in the vehicle to be inspected from the inspection object image acquired by the image acquisition unit 111. Next, in S203, the divided image generation unit 113 generates a divided image by dividing the image extracted by the symmetric portion extraction unit 112 into two in the symmetric direction.

Next, in S204, the inspection data acquisition unit 132 acquires the divided image generated by the divided image generation unit 113 as inspection data. Next, in S205, the determination unit 133 causes each inspection data item acquired by the inspection data acquisition unit 132 (that is, the divided image).
Is input to the learning model stored in the learning model storage unit 131 to determine whether the appearance of the symmetrical portion of the vehicle to be inspected is acceptable or not.

The vehicle inspection device 100 may further execute a notification process for notifying the result of the appearance inspection of the vehicle to be inspected (that is, the pass/fail judgment result) obtained by executing the above flow.

In the present embodiment, the image processing unit 110 functions as the “first image processing unit” according to the first aspect of the present invention, and the learning unit 120 functions as the “learning unit” according to the first aspect of the present invention. To do. Further, in the present embodiment, the image processing unit 110 also functions as the “second image processing unit” according to the second aspect of the present invention, and the inspection unit 130 is the “inspection unit” according to the second aspect of the present invention. Function as". That is, in the present embodiment, the vehicle inspection device 100 is configured to include the “information processing device” and the “appearance inspection device” according to the present invention. However, the “information processing device” and the “appearance inspection device” according to the present invention do not necessarily have to be configured as one device, and may be configured as different devices. Further, the “inspection object” according to the present invention is not limited to the “vehicle”, and may be any object that includes a symmetrical portion in its appearance. For example, the present invention may be applied to the appearance inspection of one component to be assembled in a vehicle.

<Other embodiments>
The above-described embodiment is merely an example, and the present invention can be appropriately modified and implemented without departing from the scope of the invention. Further, the processes and means described in the present disclosure can be freely combined and implemented as long as there is no technical contradiction.

Further, the processing described as being performed by one device may be shared and executed by a plurality of devices. Alternatively, the processing described as being performed by different devices may be performed by one device. In the computer system, it is possible to flexibly change the hardware configuration to realize each function.

100..Vehicle inspection device 110..image processing unit 111..image acquisition unit 112...symmetry part extraction unit 113..divided image generation unit 114..inverted image generation unit 120..learning unit 121..teacher data acquisition Unit 122 learning model generation unit 130 inspection unit 131 learning model storage unit 132 inspection data acquisition unit 133 discrimination unit 200 imaging device

Claims (3)

In an information processing device that generates a learning model used for appearance inspection of an inspection target by performing machine learning using a learning image obtained by capturing an object of the same type as the inspection target as teacher data,
From the learning image, while extracting an image of a symmetrical portion which is a symmetrical portion or a vertically symmetrical portion of the object, a divided image is generated by dividing the image of the symmetrical portion into two in the symmetrical direction, and further, A first image processing unit that generates an inverted image by inverting each of the image on one side and the image on the other side in the divided image;
The image on one side in the divided image and the inverted image obtained by inverting the image on the other side in the divided image are acquired as the teacher data for one side in the symmetrical portion, and the other side in the divided image is acquired. Image and the inverted image obtained by inverting the image on one side of the divided image as the teacher data for the other side of the symmetrical portion, and performing machine learning based on the acquired teacher data. And a learning unit that generates the learning model with
An information processing apparatus including. An appearance inspection apparatus that performs an appearance inspection of an inspection object using the learning model generated by the information processing apparatus according to claim 1.
From an inspection target image obtained by imaging the inspection target, an image of a symmetric portion that is a left-right symmetric or a vertical symmetric portion of the inspection target is extracted, and the symmetric portion extracted from the inspection target image A second image processing unit that generates a divided image by dividing the image of 2 into two in the symmetric direction,
An inspection unit that determines whether the appearance of the symmetrical portion of the inspection target is acceptable by acquiring the divided image generated from the inspection target image as inspection data and inputting the acquired inspection data to the learning model. When,
Appearance inspection device equipped with. In an information processing method for generating a learning model used for appearance inspection of an inspection target by performing machine learning using a learning image obtained by capturing an object of the same type as the inspection target as teacher data,
From the learning image, a step of extracting an image of a symmetrical portion which is a symmetrical portion or a vertically symmetrical portion of the object,
Generating a divided image by dividing the image of the symmetrical portion into two in the symmetric direction,
Generating an inverted image by inverting each of the image on one side and the image on the other side in the divided image;
The image on one side in the divided image and the inverted image obtained by inverting the image on the other side in the divided image are acquired as the teacher data for one side in the symmetrical portion, and the other side in the divided image is acquired. And an image obtained by inverting the image on one side in the divided image as the teacher data for the other side in the symmetrical portion,
Generating the learning model by performing machine learning based on the acquired teacher data,
An information processing method including.

Images