Inspection system, inspection method, model generation system, judgment system, model generation method, and program

Abstract

The invention addresses the problem of improving the accuracy of identifying a material to be inspected in an inspection system and an inspection method using an inspection target image obtained by imaging an inspection target and the background. This inspection system (100) comprises an inspection image acquisition unit (31), a spectrum correction unit (32), a spectrum determination unit (33), and a result output unit (40). The inspection image acquisition unit (31) acquires an inspection target image. The inspection target image is obtained by imaging a subject including an inspection target and the background in 4 or more wavelength regions. The spectrum correction unit (32) generates a second spectrum by correcting a first spectrum on the basis of the spectrum of an image of the background. The first spectrum is the spectrum of the image of the inspection target in the inspection target image. The spectrum determination unit (33) determines, on the basis of the second spectrum, whether the inspection target is a first material.

Description

[Technical field]

[0001] The present disclosure relates to an inspection system, an inspection method, a model generation system, a judgment system, a model generation method, and a program, and in particular to an inspection system, an inspection method, a model generation system, a judgment system, a model generation method, and a program that use images including data in four or more wavelength ranges. [Background technology]

[0002] As a method for distinguishing between substances that are difficult to distinguish using normal color images using the three primary colors RGB or by visual inspection with the naked eye, a method using an image containing data in many wavelength ranges is used (see, for example, Patent Document 1). A normal color image contains data in wavelength ranges corresponding to R, G, and B. In other words, a normal color image contains data in three wavelength ranges. In contrast, so-called multispectral images and hyperspectral images contain data in four or more wavelength ranges. Multispectral images and hyperspectral images contain data in wavelength ranges in the visible light region, the ultraviolet region, and four or more wavelength ranges in the infrared region. For example, the hyperspectral image disclosed in Patent Document 1 contains data in multiple wavelength ranges.

[0003] In the conventional system described in Patent Document 1, a method of identifying a substance using an image containing data in many wavelength ranges assumes that the reflection spectrum of the subject is the same as the spectrum of a pixel contained in the image of the subject in the image data, and identifies the substance. The subject includes an object to be inspected and a background. However, in an image of the object to be inspected in which the subject to be inspected and the background are captured, the image of the object to be inspected and the image of the background may not be sufficiently separated. That is, the reflection spectrum of the background may be mixed into the spectrum of a pixel contained in the image of the object to be inspected. Therefore, a difference may occur between the spectrum of the pixel contained in the image of the object to be inspected and the original spectrum of the substance of the object to be inspected, and the accuracy of identifying the substance of the object to be inspected may decrease. [Prior art documents] [Patent documents]

[0004] [Patent Document 1] JP 2010-256303 A Summary of the Invention

[0005] The present disclosure has been made in consideration of the above-mentioned points, and aims to improve the accuracy of identifying a substance of an inspection object in an inspection system, an inspection method, a model generation system, a judgment system, a model generation method, and a program that use an inspection object image in which the inspection object and background are captured.

[0006] An inspection system according to an embodiment of the present disclosure includes an inspection image acquisition unit, a spectrum correction unit, a spectrum determination unit, and a result output unit. The inspection image acquisition unit acquires an inspection object image. The inspection object image is obtained by capturing an image of a subject including an inspection object and a background in four or more wavelength ranges. The spectrum correction unit performs correction on a first spectrum based on the spectrum of an image of the background to generate a second spectrum. The first spectrum is the spectrum of an image of the inspection object in the inspection object image. The spectrum determination unit determines whether or not the inspection object is a first material based on the second spectrum. The result output unit outputs a result of the determination by the spectrum determination unit. The spectrum correction unit performs the correction on the first spectrum based on a first reference spectrum and a second reference spectrum to generate the second spectrum. The first reference spectrum is the spectrum of an image of the second material in a first reference image. The second reference spectrum is the spectrum of an image of the second material in a second reference image. The first reference image is obtained by imaging the second material, which has at least one of a size, a shape, and an arrangement known and includes a portion whose width is less than a threshold value. The second reference image is obtained by imaging the second material, which has at least one of a size, a shape, and an arrangement known and whose width is equal to or greater than the threshold value.

[0007] A model generation system according to an embodiment of the present disclosure includes a reference image acquisition unit and a correction model generation unit. The reference image acquisition unit acquires a first reference image and a second reference image. The first reference image and the second reference image are obtained by capturing a predetermined material having at least one of a size, a shape, and an arrangement known, and a background in four or more wavelength ranges. The correction model generation unit generates a spectrum correction model using teacher data based on a first reference spectrum and a second reference spectrum. The first reference spectrum is a spectrum of an image of the predetermined material in the first reference image. The second reference spectrum is a spectrum of an image of the predetermined material in the second reference image. The first reference image is obtained by capturing an image of the predetermined material having at least one of a size, a shape, and an arrangement known, and including a portion whose width is less than a threshold value. The second reference image is obtained by capturing an image of the predetermined material having at least one of a size, a shape, and an arrangement known, and whose width is equal to or greater than the threshold value. The spectrum correction model is a machine learning model for correcting a spectrum of an image of an object to be inspected based on a spectrum of an image of a background in an image of the object to be inspected, the image of the object to be inspected being obtained by capturing an image of a subject including the object to be inspected and the background in four or more wavelength bands.

[0008] A determination system according to an embodiment of the present disclosure includes a correction model acquisition unit, an inspection image acquisition unit, a spectrum correction unit, a spectrum determination unit, and a result output unit. The correction model acquisition unit acquires a spectrum correction model from the model generation system. The inspection image acquisition unit acquires an inspection object image obtained by capturing an image of a subject including an inspection object and a background in four or more wavelength ranges. The spectrum correction unit corrects a first spectrum, which is a spectrum of an image of the inspection object in the inspection object image, using the spectrum correction model to generate a second spectrum. The spectrum determination unit determines whether the inspection object is a first material or not based on the second spectrum. The result output unit outputs a result of the determination by the spectrum determination unit.

[0009] An inspection method according to one aspect of the present disclosure includes acquiring an image of an object to be inspected, correcting a first spectrum, which is a spectrum of an image of the object to be inspected in the image of the object to be inspected, based on a spectrum of an image of a background to generate a second spectrum, determining whether the object to be inspected is a predetermined material based on the second spectrum, and outputting a result of the determination. The image of the object to be inspected is obtained by capturing an image of a subject including the object to be inspected and the background in four or more wavelength bands. When generating the second spectrum, the correction is performed on the first spectrum based on a first reference spectrum and a second reference spectrum to generate the second spectrum. The first reference spectrum is the spectrum of an image of the second material in a first reference image. The second reference spectrum is the spectrum of an image of the second material in a second reference image. The first reference image is obtained by imaging the second material, which has at least one of a size, a shape, and an arrangement known and includes a portion whose width is less than a threshold value. The second reference image is obtained by imaging the second material, which has at least one of a size, a shape, and an arrangement known and whose width is equal to or greater than the threshold value.

[0010] A program according to one embodiment of the present disclosure is a program that causes one or more processors to execute the inspection method.

[0011] A model generating method according to an embodiment of the present disclosure includes acquiring a first reference image and a second reference image, and generating a spectrum correction model using teacher data based on the first reference spectrum and the second reference spectrum. The first reference image and the second reference image are obtained by capturing a predetermined material having at least one of dimensions, shape, and arrangement known, and a background in four or more wavelength ranges. The first reference spectrum is a spectrum of an image of the predetermined material in the first reference image. The second reference spectrum is a spectrum of an image of the predetermined material in the second reference image. The first reference image is obtained by capturing an image of the predetermined material having at least one of dimensions, shape, and arrangement known and including a portion whose width is less than a threshold value. The second reference image is obtained by capturing an image of the predetermined material having at least one of dimensions, shape, and arrangement known and including a portion whose width is equal to or greater than the threshold value. The spectrum correction model is a machine learning model for performing correction based on a spectrum of an image of a background in an image of an object to be inspected. The inspection object image is obtained by capturing an image of a subject including the inspection object and the background in four or more wavelength ranges.

[0012] A program according to one embodiment of the present disclosure is a program that causes one or more processors to execute the model generation method. [Brief description of the drawings]

[0013] [Figure 1] FIG. 1 is a functional block diagram of an inspection system according to an embodiment. [Diagram 2] FIG. 2 is a schematic diagram showing an image obtained by capturing an image of a first substance and a background. [Diagram 3] Fig. 3A is a graph showing the reflectance spectrum of a first material and the reflectance spectrum of a background. Fig. 3B is a graph showing the spectrum of one pixel of the image shown in Fig. 2, the reflectance spectrum of a first material, and the reflectance spectrum of a background. Fig. 3C is a graph showing the spectrum of one pixel different from Fig. 3B of the image shown in Fig. 2, the reflectance spectrum of a first material, and the reflectance spectrum of a background. [Figure 4] Fig. 4A is a schematic diagram showing an example of a subject of a first reference image, and Fig. 4B is a schematic diagram showing an example of a subject of a second reference image. [Diagram 5] Fig. 5A is a schematic diagram showing a first reference image obtained by imaging the subject of Fig. 4A. Fig. 5B is a schematic diagram showing a second reference image obtained by imaging the subject of Fig. 4B. Fig. 5C is a graph showing examples of the first and second reference spectra. [Figure 6] FIG. 6 is a schematic diagram showing the relationship between the image of an object in the first reference image and the size, shape, and location of the object. [Figure 7] FIG. 7 is a schematic diagram showing an example of a subject including an inspection target. [Figure 8] Fig. 8A is a schematic diagram showing an inspection object image obtained by imaging the subject in Fig. 7. Fig. 8B is a graph showing the relationship between a first spectrum and a second spectrum in the inspection object image in Fig. 8A. [Figure 9] FIG. 9 is a flowchart showing the operation of the above inspection system. [Figure 10] FIG. 10 is a functional block diagram of an inspection system according to the first modification. [Figure 11] FIG. 11 is a functional block diagram of a model generation system according to the second modification. [Figure 12]FIG. 12 is a schematic diagram showing an example of a subject of a teacher image. [Figure 13] Fig. 13A is a schematic diagram showing a teacher image obtained by imaging the subject of Fig. 12. Fig. 13B is a graph showing the relationship between a first spectrum and a second spectrum in the teacher image. [Figure 14] FIG. 14 is a flowchart showing the operation of the model generation system. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0014] An inspection system, an inspection method, a model generation system, a determination system, a model generation method, and a program according to embodiments will be described below with reference to the drawings.

[0015] (Embodiment) 1. Overview of the inspection system The configuration of an inspection system 100 according to an embodiment will be described with reference to the drawings.

[0016] 1.1 Components of an Inspection System 1, an inspection system 100 according to the embodiment includes a learning unit 10, a model storage unit 20, and a processing unit 30. Each of the learning unit 10 and the processing unit 30 is connected to an imaging device 2.

[0017] The inspection system 100 acquires a first reference image and a second reference image from the imaging device 2, and generates a spectral correction model. The inspection system 100 also acquires an inspection target image from the imaging device 2, performs spectral correction based on the spectral correction model, and determines the material of the subject of the inspection target image. Each of the first reference image, the second reference image, and the inspection target image is a so-called multispectral image or hyperspectral image that includes light intensity information in four or more wavelength ranges.

[0018] 1.2 Imaging device The imaging device 2 is a camera that generates an image including light intensity information of four or more non-overlapping wavelength ranges. The imaging device 2 is, for example, a spectroscopic camera equipped with a spectrometer that splits light from a subject into four or more wavelength ranges. The imaging device 2 is, for example, a so-called multispectral camera or hyperspectral camera.

[0019] The wavelength range in which the imaging device 2 detects the light intensity and reflects it in an image includes a wavelength range necessary for identifying the first material that is the inspection target of the inspection system 100. The wavelength range necessary for identifying the first material includes, for example, a peak wavelength in the absorption spectrum of the first material, or a peak wavelength in the reflection spectrum of the first material. Alternatively, the wavelength range necessary for identifying the first material is, for example, a wavelength range in which there is a large difference between the reflection spectrum of the first material and the reflection spectrum of another material that needs to be distinguished from the first material.

[0020] The wavelength ranges in which the imaging device 2 detects light intensity and stores the image may be visible light wavelengths, ultraviolet wavelengths, or infrared wavelengths. For example, when silver that may be present near aluminum is the object of inspection, the wavelength range may include a part of the wavelength range of 300 μm to 400 μm and / or a part of the wavelength range of 800 μm to 900 μm, where the difference between the reflection spectrum of aluminum and that of silver is significant.

[0021] The imaging device 2 outputs an image of a subject captured by the imaging device 2. The image includes light intensity information of each wavelength range at each pixel. Hereinafter, light intensity information of each wavelength range at a pixel will be referred to as the spectrum of that pixel.

[0022] 2. Components of the Inspection System Below, each of the components of the inspection system 100, that is, the learning unit 10, the model storage unit 20, the processing unit 30, and the result output unit 40, will be described in more detail.

[0023] 2.1 Learning section The configuration and operation of the learning unit 10 will be described in more detail below.

[0024] 2.1.1 Image Spectral Correction FIG. 2 is a schematic diagram showing an inspection object image 210 obtained by capturing an image of a subject including a first material, which is a material to be inspected, and a background. In the inspection object image 210 shown in FIG. 2, images 211 and 212 of the first material are shown in black, and an image 209 of the background is shown in white. FIG. 3A is a graph showing a reflection spectrum 311 of the first material and a reflection spectrum 312 of the background. FIG. 3B is a graph showing a spectrum 313 of a pixel included in the image 211 of the first material in the inspection object image 210 shown in FIG. 2, in comparison with the reflection spectrum 311 and the reflection spectrum 312. FIG. 3C is a graph showing a spectrum 314 of a pixel included in the image 212 of the first material in the inspection object image 210 shown in FIG. 2, in comparison with the reflection spectrum 311 and the reflection spectrum 312.

[0025] 3B and 3C, spectrum 313 of a pixel in image 211 of the first material and spectrum 314 of a pixel in image 212 of the first material do not necessarily match reflection spectrum 311 of the first material. In addition, as shown in Figures 3B and 3C, even if images 211 and 212 are both images of the first material, spectrum 313 of a pixel in image 211 and spectrum 314 of a pixel in image 212 do not necessarily match. This is because reflected light from the two types of materials is mixed at pixels near the boundary between the images of the two types of materials in inspection object image 210.

[0026] That is, the spectrum of a pixel at or near the boundary between the first material image 211 and the background image 209 is a mixture of the first material reflection spectrum 311 and the background reflection spectrum 312. The degree of mixing of the first material reflection spectrum 311 and the background reflection spectrum 312 varies depending on the size, shape, and arrangement of the first material.

[0027] For example, the smaller the size of the first material, the greater the spectral mixing at and near the boundary between the first material images 211, 212 and the background image 209. This is because spectral mixing is stronger closer to the boundary between the first material images 211, 212 and the background image 209, and therefore when the size of the first material is small, the area of ​​the region where spectral mixing occurs is larger compared to the area occupied by the first material images 211, 212.

[0028] For the same reason, the degree of spectral mixing differs depending on the shape of the first material. That is, the longer the boundary between the first material images 211, 212 and the background image 209, the greater the effect of spectral mixing at the boundary between the first material images 211, 212 and the background image 209 and the pixels in the vicinity thereof. Therefore, for example, the greater the ratio of the length in the major axis direction to the length in the minor axis direction in the shape of the first material, the greater the spectral mixing at the boundary between the first material images 211, 212 and the background image 209 and the pixels in the vicinity thereof.

[0029] Furthermore, when light is incident from a direction other than perpendicular to the background, the shape and position of the area in the inspection target image 210 where spectral mixing occurs will vary depending on the angle between the long axis direction of the shape of the first material and the direction of incidence of the light.

[0030] In view of the above reasons, the learning unit 10 creates a correction model for reducing the influence of the spectrum of the background from the spectrum of each pixel in the inspection object image 210 obtained by capturing an image of a subject including the inspection object.

[0031] 2.1.2 Structure and Function of the Learning Unit Each component of the learning unit 10 will now be described.

[0032] As shown in Fig. 1, the learning unit 10 includes a reference image acquisition unit 11 and a correction model generation unit 12. The reference image acquisition unit 11 is, for example, a combination of a processor, a memory, and a program. The correction model generation unit 12 is, for example, a combination of a processor, a memory, and a program. The processor of the reference image acquisition unit 11 and the processor of the correction model generation unit 12 may be the same processor. The processor is, for example, a CPU (Central Processing Unit) or a GPU (Graphics Processing Unit).

[0033] (1) Overview of the reference image acquisition unit The reference image acquisition unit 11 acquires a first reference image, a second reference image, and a third reference image that satisfy the following conditions. Hereinafter, the first reference image, the second reference image, and the third reference image will be referred to as “reference images” when referring to them collectively. The reference image acquisition unit 11 acquires information necessary for generating a correction model from the reference images, and outputs the information to the correction model generation unit 12.

[0034] (1.1) Details of the first reference image The reference image acquisition unit 11 acquires a first reference image 421 that satisfies the following conditions from the imaging device 2. The first reference image 421 (see FIG. 5A) is an image of a subject 411 (see FIG. 4A) including a second material 213 and a background 215, captured by the imaging device 2. The first reference image 421 is used to create training data for creating a correction model. The second material is a material used to create training data for creating a correction model. The second material is, for example, the same material as the first material. Note that, as described later, the second material may be a material different from the first material.

[0035] The first reference image 421 also includes one or more images of an object 213 (see FIG. 4A) made of a second material. The object 213 has a length (width) in the short axis direction less than a threshold. The threshold is determined as follows. In an image captured of a state in which an object made of a second material and having a length (width) in the short axis direction equal to or greater than a threshold is adjacent to a background, the degree of mixing of the reflection spectrum of the background in the spectrum of a pixel included in the image of the object is set as a first degree of mixing. In an image captured of a state in which an object made of a second material and having a length (width) in the short axis direction less than a threshold is adjacent to a background, the degree of mixing of the reflection spectrum of the background in the spectrum of a pixel included in the image of the object is set as a second degree of mixing. The second degree of mixing is greater than the first degree of mixing. The threshold is, for example, 1 mm.

[0036] The first reference image 421 may include a plurality of images of the object 213. FIG. 4A shows an example of the object 411 of the first reference image 421 (see FIG. 5A). The object 411 includes the objects 213-1 to 213-25, each of which satisfies the condition of the object 213 described above. FIG. 5A shows the first reference image 421 obtained by capturing the object 411 (see FIG. 4A). The first reference image 421 includes a plurality of images 221-1 to 221-25. The plurality of images 221-1 to 221-25 are the images of the objects 213-1 to 213-25, respectively. Hereinafter, when the objects 213-1 to 213-25 are not distinguished from one another, they will be referred to as "objects 213". When the images 221-1 to 221-25 are not distinguished from one another, they will be referred to as "images 221".

[0037] Note that the shape of the object 213 is not limited as long as the object 213 is made of a second material and has a length (width) in the minor axis direction less than a threshold value. The object 213 may be, for example, a particle of the second material as shown in Fig. 4A. Alternatively, for example, the object 213 may be a TEG (Test Element Group) pattern including a second material whose length (width) in the minor axis direction is less than a threshold value.

[0038] Furthermore, parameters indicative of the size, shape, and location of the object 213 in the first reference image 421 are known.

[0039] The parameter indicating the dimension is, for example, the maximum diagonal distance. The maximum diagonal distance is the length of the maximum diagonal. As shown in FIG. 6, the maximum diagonal 231 is the line segment with the longest length among the line segments connecting any two points existing on the boundary between the object 213 and the background 215 in a plan view from the imaging device 2. For example, when the shape of the object 213 is an ellipse in a plan view from the imaging device 2, the maximum diagonal is the major axis of the ellipse. Or, for example, when the shape of the object 213 is a parallelogram in a plan view from the imaging device 2, the maximum diagonal is the diagonal with the longer length among the two diagonals of the parallelogram.

[0040] The parameter indicating the shape is, for example, an area ratio obtained by dividing the area of ​​the object 213 by the square of the maximum diagonal distance in a planar view from the imaging device 2. For example, when the shape of the object 213 is a square in a planar view from the imaging device 2, the area ratio is 50%. Also, for example, when the shape of the object 213 is a rectangle with an aspect ratio of 3:1 in a planar view from the imaging device 2, the area ratio is 30%. The area ratio decreases as the ratio of the perimeter to the area of ​​the shape of the object 213 increases in a planar view from the imaging device 2.

[0041] The parameter indicating the arrangement is, for example, a value indicating the orientation of the maximum diagonal 231 relative to the orientation of the first reference image 421. For example, when the angle θ between the direction of the maximum diagonal and the direction that is the x-axis of the first reference image 421 is 20°, the value indicating the orientation is 20°.

[0042] (1.2) Processing for the first reference image The reference image acquisition unit 11 acquires the first reference image 421 that satisfies the above-mentioned conditions. Furthermore, the reference image acquisition unit 11 acquires parameters indicating the dimensions, shape, and arrangement of the object 213 for each of the objects 213 that are the subjects of the first reference image 421. For example, the reference image acquisition unit 11 acquires parameters indicating the dimensions, shape, and arrangement of the object 213 from a database (not shown), which parameters were acquired in advance when the object 213 was prepared. Alternatively, the reference image acquisition unit 11 may calculate the parameters indicating the dimensions, shape, and arrangement of the object 213 by image processing, as described later.

[0043] The reference image acquiring unit 11 acquires the spectrum data of the pixels included in the image 221 of the first reference image 421 as the first reference spectrum 321 (see FIG. 5C). The reference image acquiring unit 11 outputs the first reference spectrum 321 for each image 221 to the correction model generating unit 12 in association with the size, shape, and arrangement of the object 213. Note that, for the first reference image 421 including a plurality of images 221, the reference image acquiring unit 11 outputs the first reference spectrum 321 of the image 221 for each image 221 to the correction model generating unit 12 in association with the size, shape, and arrangement of the object 213. Furthermore, when the reference image acquiring unit 11 acquires a plurality of first reference images 421, the above-mentioned process is performed for each first reference image 421.

[0044] (1.3) Method for calculating parameters that indicate the size, shape, and arrangement of an object The reference image acquisition unit 11 may perform image processing on the first reference image 421 and calculate parameters indicating the size, shape, and arrangement of the object 213 using the resolution of the first reference image 421. The image processing is, for example, as follows.

[0045] The reference image acquisition unit 11 extracts the image 221 from the first reference image 421 by, for example, edge extraction processing. Next, the reference image acquisition unit 11 detects the maximum diagonal image of the image 221. The maximum diagonal image is a line segment with the longest length among line segments connecting any two points on the periphery of the image 221. Since the maximum diagonal image is an image of the maximum diagonal 231, the reference image acquisition unit 11 calculates the maximum diagonal distance based on the length of the maximum diagonal image and the resolution of the first reference image 421.

[0046] Further, the reference image acquisition unit 11 calculates a parameter indicating the shape as follows. For example, the reference image acquisition unit 11 measures the number of pixels included in the image 221, and divides the measured number of pixels by the square of the length in pixels of the maximum diagonal image to calculate the area ratio of the image 221. The number of pixels included in the image 211 of the object 213 is proportional to the area of ​​the object 213. Moreover, the square of the length of the maximum diagonal image is proportional to the square of the length of the maximum diagonal distance. Furthermore, the ratio of the number of pixels included in the image 211 of the object 213 to the area of ​​the object 213 is equal to the ratio of the square of the length of the maximum diagonal image to the square of the length of the maximum diagonal distance. Therefore, the area ratio of the image 211 is equal to the area ratio of the object 213. Instead of the area ratio of the object 213, the reference image acquisition unit 11 calculates the area ratio of the image 211, which is the same value.

[0047] Further, the reference image acquiring unit 11 calculates parameters indicating the arrangement as follows. For example, the reference image acquiring unit 11 calculates the coordinates on the two first reference images 421 that are both ends of the maximum diagonal image. Then, the reference image acquiring unit 11 divides the difference between the y coordinates of the two coordinates on the first reference image 421 by the length of the maximum diagonal image, and calculates the sine (sin θ) value of the angle θ between the direction of the maximum diagonal image and the x-axis of the first reference image 421. The reference image acquiring unit 11 calculates the angle θ between the direction of the maximum diagonal image and the x-axis of the first reference image 421 from the calculated sine value sin θ.

[0048] (1.4) Second and third reference images The reference image acquisition unit 11 acquires the second reference image 422 that satisfies the following conditions from the imaging device 2. FIG. 4B shows an example of the subject 412 of the second reference image 422. The subject 412 includes the object 214. As shown in FIG. 5B, the second reference image 422 includes an image 222 of the object 214. The object 214 has a length (width) in the short axis direction of the object 214 equal to or greater than a threshold value in a planar view from the imaging device 2. That is, the width of the object 214 is always equal to or greater than a threshold value in a planar view from the imaging device 2. As described above, an object whose short axis direction length is equal to or greater than a threshold value is less influenced by the spectrum of the background than an object whose short axis direction length is less than the threshold value. That is, the influence of the spectrum of the background 215 in the spectrum of the image 222 of the object 214 is less than the influence of the spectrum of the background 215 in the image 221 of the first reference image 421.

[0049] Note that the object 214 may have a length (width) in the minor axis direction of the object 214 equal to or greater than the threshold value when viewed in a plan view from the imaging device 2. Therefore, the object 214 may be, for example, a flat plate that includes a square with one side equal to the threshold value when viewed in a plan view from the imaging device 2. That is, the object 214 may be a bulk like the subject 412 shown in FIG. 4B. Note that the object 214 may be particulate or a TEG pattern as long as the length (width) in the minor axis direction of the object 214 is equal to or greater than the threshold value.

[0050] The reference image acquisition section 11 acquires the spectral data of the pixels included in the image 222 of the object 214 included in the second reference image 422 as the second reference spectrum 322 (see FIG. 5C).

[0051] The reference image acquiring unit 11 acquires a third reference image that satisfies the following conditions from the imaging device 2. In this embodiment, as shown in Figs. 4A and 5A, a first reference image 421 in which a subject 411 including a background 215 is captured is also used as the third reference image. The third reference image includes an image 223 of the background 215. The reference image acquiring unit 11 extracts spectral data of pixels included in the image 223 of the background 215 in the third reference image as a reflection spectrum 323 of the background (see Fig. 5C).

[0052] (2) Overview of the correction model generation part The correction model generating unit 12 estimates the influence of the reflection spectrum of the background in the spectral data by the following procedure. The first reference spectrum 321 of a pixel included in the image 221-n of any object 213-n in the first reference image 421 is taken as spectrum A1(n). The index n is an index indicating which object 213 the image indicates. The second reference spectrum 322 is taken as spectrum A2. Furthermore, the reflection spectrum 323 of the background 215 is taken as spectrum A3. The value α that satisfies the following equation is taken as the influence α of the background spectrum in the object 213-n.

[0053] A1(n)=α×A3+(1-α)×A2…(Formula 1)

[0054] Therefore, the value of the influence α is calculated by the following formula:

[0055] α={A1(n)-A2} / {A3-A2}…(Formula 2)

[0056] The correction model generating unit 12 associates the value of the influence α calculated for each index n with parameters indicating the dimensions, shape, and arrangement of the object 213-n. Then, the correction model generating unit 12 creates training data for machine learning in which the dimensions, shape, and arrangement of the object 213 are input values ​​and the value of the influence α is correct data. The correction model generating unit 12 performs machine learning using the created training data to generate a correction model. For example, a support vector machine or a random forest can be used as the machine learning method. Note that the machine learning method may be any method as long as it is a learning method using training data, and may be, for example, a neural network. By machine learning using the training data, a correction model is generated that inputs the dimensions, shape, and arrangement of the object 213 and outputs the influence α of the background spectrum in the object 213.

[0057] The correction model generating unit 12 outputs the generated correction model to the correction model holding unit 21.

[0058] 2.2 Model Storage The model storage unit 20 includes a correction model holding unit 21 and a determination model holding unit 22 .

[0059] The correction model holding unit 21 is a storage medium that holds the correction model generated by the above-mentioned correction model generation unit 12. Note that the correction model holding unit 21 may hold a correction model acquired from another system. The other system includes, for example, the correction model generation unit 12. Or, the other system includes, for example, the correction model holding unit 21. That is, the correction model holding unit 21 holds the correction model generated by the correction model generation unit 12 or the correction model generated by the other system.

[0060] The judgment model holding unit 22 is a storage medium that holds a judgment model. As described later, the judgment model is a model for identifying a material of the inspection object 216 (see FIG. 7) based on a spectrum obtained by correcting a spectrum of a pixel included in an image 224 of the inspection object in an inspection object image 420 (see FIG. 8) obtained by capturing the inspection object 216 (see FIG. 7).

[0061] 2.3 Processing section 1, the processing unit 30 includes an inspection image acquisition unit 31, a spectrum correction unit 32, and a spectrum determination unit 33. Each of the inspection image acquisition unit 31, the spectrum correction unit 32, and the spectrum determination unit 33 is realized, for example, by a computer having a processor and a program. Note that any two or all of the inspection image acquisition unit 31, the spectrum correction unit 32, and the spectrum determination unit 33 may be realized using a single processor or a single computer.

[0062] (1) Inspection image acquisition unit The inspection image acquisition unit 31 acquires an inspection object image 420 including an image 224 of the inspection object 216 and an image 225 of the background 217 from the imaging device 2 as shown in Fig. 8A. The inspection object image 420 is, for example, an image obtained by capturing an image of a subject 410 shown in Fig. 7. The subject 410 includes the inspection object 216 and the background 217. The inspection object image 420 includes an image 224 of the inspection object 216 and an image 225 of the background 217.

[0063] The inspection image acquisition unit 31 extracts one or more images 224 of the inspection object 216 and an image 225 of the background 217 by, for example, edge extraction. For each of the images 224 of the inspection object 216, the inspection image acquisition unit 31 calculates a parameter indicating the size, a parameter indicating the shape, and a parameter indicating the arrangement of the image 224 by the same method as the reference image acquisition unit 11. The inspection image acquisition unit 31 acquires a spectrum of a pixel included in the image 224 of the inspection object 216 included in the inspection object image 420 as a first spectrum 315 (see FIG. 8B). For each of the images 224 of the inspection object 216 included in the inspection object image 420, the inspection image acquisition unit 31 outputs the first spectrum 315 to the spectrum correction unit 32 in association with the parameters indicating the size, the shape, and the arrangement of the image 224. Furthermore, inspection image acquisition unit 31 outputs spectrum 316 of a pixel included in image 225 of background 217 in image 420 to be inspected to spectrum correction unit 32 as a reflection spectrum of the background. Furthermore, inspection image acquisition unit 31 outputs the entire image 420 to be inspected to spectrum correction unit 32.

[0064] (2) Spectral correction section The spectrum correction unit 32 performs correction on the first spectrum 315 of a pixel included in the image 224 of the inspection object 216 in the inspection object image 420 based on the correction model to reduce the influence of the background reflection spectrum 316, thereby generating a second spectrum 317.

[0065] The spectrum correction unit 32 acquires the correction model from the correction model storage unit 21. The spectrum correction unit 32 estimates the value of the influence α of the background spectrum for each image 224 of the inspection object 216 in the inspection object image 420 based on a combination of parameters indicating the size, shape, and arrangement of the image 224 and the correction model. This process calculates the value of the influence α of the background spectrum in the image 224 based on the assumption that if the features of the size, shape, and arrangement of the subject are similar between the first reference image 421 and the inspection object image 420, the influence α of the background spectrum is also similar. The first spectrum 315 of the pixel included in the image 224 of the inspection object 216 in the inspection object image 420 is set as spectrum A4(m). The index m is an index indicating which of the multiple images 224 is indicated. The spectrum A3 is the spectrum 316 of the image 225 of the background 217. Furthermore, if the second spectrum 317 in which the influence of the background spectrum is reduced is taken as spectrum A5, the following equation holds, similar to equation 1.

[0066] A4(m)=α×A3+(1-α)×A5…(Formula 3)

[0067] Therefore, the second spectrum 317 can be calculated by the following formula:

[0068] A5={A4(m)-α×A3} / (1-α)…(Formula 4)

[0069] The spectrum correction unit 32 calculates the influence α of the background spectrum and calculates a spectrum A5 which is a second spectrum 317 for all pixels included in the image 224 of any one of the inspection objects 216 in the inspection object image 420. By this process, as shown in Fig. 8B, for the image 224 of the inspection object image 420, a corrected second spectrum 317 is calculated based on the first spectrum 315 of the pixel included in the image 224 and the spectrum 316 of the pixel included in the image 225 of the background 217. The second spectrum 317 has a low influence of the spectrum of the background to the same extent as the second reference spectrum 322 of the image 222 included in the second reference image 422.

[0070] The spectrum correction unit 32 replaces the first spectrum 315 with the second spectrum 317 for all pixels included in the image 224 of any one of the inspection objects 216 in the inspection object image 420 to generate a corrected image. The spectrum correction unit 32 outputs the generated corrected image to the spectrum determination unit 33.

[0071] The spectrum correction unit 32 may acquire a correction model from another system. The other system may have, for example, the correction model holding unit 21. Alternatively, the spectrum correction unit 32 may include an internal storage unit that holds the correction model, and may hold the correction model acquired from the correction model holding unit 21 or another system.

[0072] (3) Spectral evaluation unit The spectrum determining unit 33 identifies the material of the inspection object 216 based on the second spectrum 317 of the pixel included in the image 224 of the inspection object 216 for the corrected image.

[0073] The spectrum determination unit 33 acquires the corrected image from the spectrum correction unit 32. The spectrum determination unit 33 acquires the second spectrum 317 of the pixel included in the image 224 of the inspection object 216 from the corrected image. The spectrum determination unit 33 acquires the determination model from the determination model storage unit 22. The determination model is a model for determining the substance of the inspection object 216 based on the second spectrum 317 of the pixel included in the image 224 of the inspection object 216. The determination model is, for example, a machine learning model generated in advance using teacher data in which the reflection spectrum of a known substance is used as an input value and information identifying the substance is used as a correct answer. The information identifying the substance is, for example, the name or abbreviation of the substance. Alternatively, the information identifying the substance may be, for example, information indicating whether or not the substance matches a predetermined substance. The spectrum determination unit 33 determines the substance of the inspection object 216 from the determination model and the second spectrum 317 of the pixel included in the image 224 of the inspection object 216.

[0074] 2.4 Result output section The result output unit 40 is a device that displays to a user the determination result of the material of the inspection object 216, the image of which 224 is included in the inspection object image 420. The result output unit 40 is, for example, a display device.

[0075] For example, the result output unit 40 displays the image 224 of the inspection target 216 in the inspection target image 420 in a different color depending on the type of material of the inspection target 216. Also, for example, the result output unit 40 may display the second spectrum for each pixel of the inspection target image 420. For example, the result output unit 40 may be a touch panel, and when any pixel of the inspection target image 420 is touched, the second spectrum 317 of the touched pixel may be displayed. Alternatively, for example, the result output unit 40 may be a touch panel, and when any pixel of the inspection target image 420 is touched, the first spectrum 315 and the second spectrum 317 of the inspection target image 420 of the touched pixel may be displayed in a comparable manner. Also, for example, the result output unit 40 may display one or both of the first reference spectrum 321 and the second reference spectrum 322 of a pixel included in the image of the object 213 in the first reference image 421.

[0076] The result output unit 40 is not limited to a display device, and may be, for example, a database or a storage medium that stores the determination result of the spectrum determination unit 33 on the substance of the test object 216.

[0077] 3. Operation of the Inspection System The following describes the operation of the inspection system 100. Fig. 9 is a flowchart showing the operation of the inspection system 100 according to the embodiment.

[0078] The inspection system 100 acquires a reference image (step S1). The reference image acquisition unit 11 of the inspection system 100 acquires each of the first reference image 421, the second reference image 422, and the third reference image that satisfy the above-mentioned conditions. The reference image acquisition unit 11 acquires an image including the image 221 of the object 213 as the first reference image 421. In addition, in order to improve the accuracy of the correction model, it is preferable that the number of images 221 of the object 213 that can be acquired from the first reference image 421 is multiple. Therefore, the reference image acquisition unit 11 acquires multiple first reference images 421 including the image 221 of the object 213. Or, the reference image acquisition unit 11 acquires the first reference image 421 including multiple images 221 of the object 213. In addition, the reference image acquisition unit 11 acquires an image including the image 222 of the object 214 as the second reference image 422.

[0079] The reference image acquisition unit 11 may acquire an image including the image 221 of the object 213 and the image 222 of the object 214 as the first reference image 421 and may also acquire it as the second reference image 422. The reference image acquisition unit 11 acquires an image including the image 223 of the background 215 as the third reference image. The reference image acquisition unit 11 may acquire an image including the image 221 of the object 213 and the image 223 of the background 215 as the first reference image 421 and may also acquire it as the third reference image. Alternatively, the reference image acquisition unit 11 may acquire an image including the image 222 of the object 214 and the image 223 of the background 215 as the second reference image 422 and may also acquire it as the third reference image. Alternatively, the reference image acquisition section 11 may acquire images including the image 221 of the object 213, the image 222 of the object 214, and the image 223 of the background 215 as the first reference image 421, the second reference image 422, and the third reference image.

[0080] The reference image acquisition unit 11 acquires each of the reference images, for example, from the imaging device 2. Alternatively, the reference image acquisition unit 11 may acquire the reference images from an image storage system that stores images captured by the imaging device 2.

[0081] Next, the inspection system 100 extracts information of the reference image (step S2). The reference image acquisition unit 11 of the inspection system 100 identifies the images 221 of the multiple objects 213 from one or more first reference images 421. The reference image acquisition unit 11 acquires parameters indicating the dimensions, shape, and arrangement of the object 213 for each of the multiple objects 213. The reference image acquisition unit 11 performs the above-mentioned image processing on the first reference image 421 including the image 221 of the object 213 as described above, and calculates the parameters indicating the dimensions, shape, and arrangement of the object 213 using the resolution of the first reference image 421. Alternatively, for example, the reference image acquisition unit 11 may accept an input from a user regarding one or more of the parameters indicating the dimensions, shape, and arrangement of the object 213. Alternatively, for example, the reference image acquisition unit 11 may acquire one or more of the parameters indicating the dimensions, shape, and arrangement of the object 213 from an image storage system together with the reference image.

[0082] The reference image acquisition unit 11 associates a first reference spectrum 321, which is a spectrum of a pixel included in an image 221 of the object 213, with parameters indicating the size, shape, and arrangement of the object 213 for each of the multiple objects 213, and outputs the first reference spectrum 321 to the correction model generation unit 12. The reference image acquisition unit 11 also outputs the spectrum of a pixel included in an image 222 of the object 214 in the second reference image 422 to the correction model generation unit 12 as a second reference spectrum 322. The reference image acquisition unit 11 also outputs the spectrum of a pixel included in an image 223 of the background 215 in the third reference image to the correction model generation unit 12. The inspection system 100 may output a known reflection spectrum of the second material to the correction model generation unit 12 instead of the second reference spectrum 322. Furthermore, inspection system 100 may output a known reflectance spectrum of background 215 to correction model generating unit 12 instead of the spectrum of the pixel of image 223 of background 215 in the third reference image.

[0083] Next, the inspection system 100 generates a correction model (step S3). The correction model generation unit 12 of the inspection system 100 calculates the influence α of the background spectrum of the first reference spectrum 321 for each of the objects 213. The correction model generation unit 12 calculates the influence α of the background spectrum using the second reference spectrum 322, the spectrum of the pixel of the image 223 of the background 215 in the third reference image, and the first reference spectrum 321. Then, for each of the multiple objects 213, the correction model generation unit 12 associates the influence α of the background spectrum of the first reference spectrum 321 with parameters indicating the size, shape, and arrangement of the object 213. By this process, teacher data for machine learning is generated, in which the parameters indicating the size, shape, and arrangement of the object 213 are used as input values ​​and the influence α of the background spectrum is used as correct answer data. The correction model generation unit 12 performs machine learning using the teacher data to generate a correction model. The correction model generating unit 12 outputs the generated correction model to the correction model holding unit 21.

[0084] Next, the inspection system 100 acquires the inspection object image 420 (step S4). The inspection image acquisition unit 31 of the inspection system 100 acquires an image including the image 224 of the inspection object 216 from the imaging device 2 as the inspection object image 420.

[0085] Next, the inspection system 100 extracts information from the inspection object image 420 (step S5). The inspection image acquisition unit 31 of the inspection system 100 extracts one or more images 224 of the inspection object 216 and an image 225 of the background 217 from the inspection object image 420. For each of the inspection objects 216, the inspection image acquisition unit 31 calculates a parameter indicating the size, a parameter indicating the shape, and a parameter indicating the arrangement of the inspection object 216 based on the image 224 of the inspection object 216 by a method similar to that of the reference image acquisition unit 11. For each of the inspection objects 216, the inspection image acquisition unit 31 outputs a combination of the parameters indicating the size, shape, and arrangement of the inspection object 216 and a first spectrum 315 of a pixel included in the image 224 of the inspection object 216 to the spectrum correction unit 32. Furthermore, inspection image acquisition unit 31 outputs the spectrum of pixels included in image 225 of background 217 of image 420 to be inspected to spectrum correction unit 32 as spectrum 316 of image 225 of background 217. Furthermore, inspection image acquisition unit 31 outputs the entire image 420 to spectrum correction unit 32.

[0086] Next, the inspection system 100 corrects the spectrum (step S6). The spectrum correction unit 32 of the inspection system 100 acquires the correction model from the correction model storage unit 21. The spectrum correction unit 32 estimates the influence α of the background spectrum in the first spectrum 315 from parameters indicating the size, shape, and arrangement of the inspection object 216 for each of the inspection objects 216 in the inspection object image 420 using the correction model. The spectrum correction unit 32 corrects the first spectrum 315 of the pixel included in the image 224 of the inspection object 216 for each of the inspection objects 216 based on the estimated influence α of the background spectrum to generate the second spectrum 317. The second spectrum 317 is generated based on the assumption that the spectrum 316 of the background affects the second spectrum 317 by the influence α of the background spectrum, thereby forming the first spectrum 315 of the pixel included in the image 224 of the inspection object 216. That is, the second spectrum 317 is a spectrum obtained by reducing the influence of the background spectrum 316 from the first spectrum 315 of the pixel included in the image 224 to the same extent as the second reference spectrum.

[0087] The spectrum correction unit 32 generates a corrected image for the inspection object image 420, which is an image in which the first spectrum 315 of all pixels included in the image 224 of the inspection object 216 is replaced with the second spectrum 317. The spectrum correction unit 32 outputs the corrected image to the spectrum determination unit 33.

[0088] Next, the inspection system 100 identifies the inspection target 216 (step S7). The spectrum determination unit 33 of the inspection system 100 identifies the material of the inspection target 216 based on the second spectrum 317 of the pixel included in the image 224 of the inspection target 216 for the corrected image. As described above, the second spectrum 317 is a spectrum in which the influence of the background spectrum 316 is reduced to the same extent as the first reference spectrum. Therefore, the second spectrum 317 is closer to the reflection spectrum of the material constituting the inspection target 216 than the first spectrum 315. The method of identifying the material is, for example, a method using a determination model by machine learning. Alternatively, the method of identifying the material may be, for example, a comparison with the reflection spectrum of a known material.

[0089] Through the above process, the material of the test object 216 is determined.

[0090] 4.Effects The inspection system 100 according to the embodiment judges the material of the inspection target 216 based on an image including light intensity information in four or more wavelength ranges. One or more of the four or more wavelength ranges may be included in either the ultraviolet range or the infrared range. Therefore, it is possible to inspect materials that are difficult to distinguish visually, such as judging a metal material that exists near a different metal, or a resin material that exists near a different resin.

[0091] Furthermore, in the inspection system 100 according to the embodiment, the spectrum correction unit 32 corrects the first spectrum 315 of the pixel included in the image 224 of the inspection object 216 in the inspection object image 420 based on the correction model. Therefore, the spectrum determination unit 33 can determine the material of the inspection object 216 based on the second spectrum 317 in which the influence of the spectrum 316 of the background is reduced. Therefore, it is possible to prevent a decrease in accuracy in determining the material of the inspection object 216 due to mixing of the spectrum 316 of the background with the first spectrum 315 of the image 224 of the inspection object 216. Therefore, it is possible to improve the accuracy in determining the material of the inspection object 216.

[0092] Moreover, the learning unit 10 according to the embodiment receives as input a combination of parameters indicating the size, shape, and arrangement of the object 213, and generates a correction model that outputs the background spectrum influence α. That is, for the first reference image 421 and the inspection target image 420, the value of the background spectrum influence α in the image 224 of the inspection target 216 is calculated based on the assumption that if the size, shape, and arrangement of the subject are similar, then the background spectrum influence α will be similar. Therefore, by preparing a wide variety of teacher data for generating the correction model, the accuracy of the correction can be easily improved, and the accuracy of determining the material of the inspection target 216 can be improved.

[0093] In addition, in the inspection system 100 according to the embodiment, the influence α of the background spectrum in the first reference spectrum 321 of the image 221 of the object 213 in the first reference image 421 is indicated by the ratio of the spectrum of the background to the first reference spectrum 321. That is, the influence α of the background spectrum does not depend on the characteristics or intensity of the reflection spectrum 311 of the first material. Therefore, if it can be estimated that the influence α of the background spectrum is the same, the second material that is the subject of the first reference image 421 and the first material that is the subject of the inspection target image 420 and is the object of identification of the inspection target 216 are not necessarily the same material. For example, when the background material is a resin and both the second material and the first material are metal materials, it can be estimated that the influence α of the background spectrum is the same even if the first material and the second material are different metals. That is, for example, by using a correction model generated using iron as the second material, it is considered that the influence of the reflection spectrum of the background can be reduced by the spectrum correction by the spectrum correction unit 32 even when the first material is silver. Therefore, it is not necessary to generate a correction model for each type of first substance in which the same type of substance is used as the second substance, and the generation of the correction model can be simplified.

[0094] Furthermore, in the inspection system 100 according to the embodiment, the spectrum determination unit 33 can determine the material of the inspection target 216 based on the second spectrum 317 in which the influence of the background spectrum 316 is reduced. Therefore, when creating a judgment model by machine learning, the judgment accuracy can be easily improved by using a spectrum of a sample in which the influence of the background spectrum 316 is similarly small and the width in the minor axis direction is equal to or greater than a threshold. Furthermore, when creating training data for creating a judgment model, it is not necessary to use a spectrum of a sample including a portion in which the width in the minor axis direction is less than a threshold, so that the generation of the judgment model can be reduced in labor. Furthermore, by not using training data including a spectrum that is greatly influenced by the background spectrum 316 as an input, it is possible to prevent the accuracy of the judgment model from decreasing.

[0095] In the spectrum determination unit 33 according to the embodiment, the second spectrum 317 of each pixel included in the image of the inspection object 216 in the corrected image is less affected by the spectrum 316 of the background. Therefore, the determination model is not limited to one formed by machine learning. For example, the determination model holding unit 22 holds the reflection spectrum of a known substance. The spectrum determination unit 33 calculates the degree of agreement between the second spectrum 317 of each pixel included in the image of the inspection object 216 in the corrected image and the reflection spectrum of the known substance held in the determination model holding unit 22. If the degree of agreement is equal to or greater than a predetermined threshold, the spectrum determination unit 33 determines that the inspection object 216 is a known substance. The degree of agreement can be determined by, for example, correlation processing. The substance of the inspection object 216 can also be identified by such processing.

[0096] (Variation 1) As shown in FIG. 10, an inspection system 130 according to the first modification differs from the inspection system 100 (see FIG. 1) according to the embodiment in that the inspection system 130 includes a determination system 110 and a model generation system 120.

[0097] As shown in FIG. 10, the model generation system 120 includes a learning unit 10 and a model storage unit 23. The learning unit 10 has the same function as the learning unit 10 (see FIG. 1) of the inspection system 100 according to the embodiment. Therefore, detailed description of the learning unit 10 will be omitted. The model storage unit 23 includes a correction model holding unit 21. The correction model holding unit 21 has the same function as the correction model holding unit 21 (see FIG. 1) of the inspection system 100 according to the embodiment. Therefore, detailed description of the correction model holding unit 21 will be omitted. That is, the model generation system 120 has the same configuration as the inspection system 100 except for the processing unit 30 and the judgment model holding unit 22. The model generation system 120 performs an operation of generating a correction model using the first reference image 421, the second reference image 422, and the third reference image in the inspection system 100. That is, the model generation system 120 performs the operations from steps S1 to S3 in the flowchart of FIG. 9.

[0098] As shown in FIG. 10, the judgment system 110 includes a processing unit 30, a result output unit 40, a correction model acquisition unit 50, and a model storage unit 24. The result output unit 40 has the same function as the result output unit 40 (see FIG. 1) of the inspection system 100 according to the embodiment. Therefore, detailed description of the result output unit 40 is omitted. The processing unit 30 has the same function as the processing unit 30 (see FIG. 1) of the inspection system 100, except that the spectrum correction unit 32 acquires the correction model from the correction model acquisition unit 50 instead of the correction model storage unit 21. The model storage unit 24 includes a judgment model storage unit 22. The judgment model storage unit 22 has the same function as the judgment model storage unit 22 (see FIG. 1) of the inspection system 100. The correction model acquisition unit 50 is a functional block that acquires a correction model from the correction model storage unit 21 of the model generation system 120, and is, for example, a communication device that can communicate with the correction model storage unit 21 of the model generation system 120. That is, the determination system 110 has a configuration in which the learning unit 10 and the correction model holding unit 21 are removed from the inspection system 100, and a correction model acquisition unit 50 is added. The determination system 110 performs an operation of correcting the spectrum of the inspection object image 420 using the correction model in the inspection system 100, and determining the material of the inspection object 216. That is, the determination system 110 performs the operations from steps S4 to S7 in the flowchart of FIG.

[0099] As described above, in the inspection system 130 according to the first modification, the model generation system 120 generates a correction model, and the determination system 110 uses the correction model to determine the material of the inspection target 216. That is, the operation of the inspection system 100 according to the embodiment is performed by the determination system 110 and the model generation system 120 in cooperation with each other.

[0100] This modification can also provide the same effects as the inspection system 100 according to the embodiment. Moreover, according to this modification, for example, a correction model formed by one model generation system 120 can be used by a plurality of determination systems 110. Therefore, it is not necessary to create a correction model for each determination system 110, and the generation of the correction model can be simplified.

[0101] In addition, in the first modification, the model generation system 120 has a correction model holding unit 21, and the determination system 110 has a correction model acquisition unit 50. Instead of this configuration, a configuration can be adopted in which the model generation system 120 has a correction model transmission unit, and the determination system 110 has the correction model holding unit 21. Or, for example, a configuration can be adopted in which the inspection system 130 includes a model storage system, the model generation system 120 outputs a correction model to the model storage system, and the determination system 110 acquires the correction model from the model storage system. The model storage system is, for example, a server on a network. Or, the model storage system is, for example, a storage medium.

[0102] (Variation 2) 1. Configuration As shown in FIG. 11, a model generation system 150 according to the second modification differs from the model generation system 120 according to the first modification (see FIG. 10) in that the model generation system 150 further includes a determination model generation system 140.

[0103] The model generation system 150 according to the modified example 2 includes a correction model generation system 120 and a judgment model generation system 140. The correction model generation system 120 is the same as the model generation system 120 according to the modified example 1. Therefore, a detailed description will be omitted.

[0104] As shown in FIG. 11, the determination model generating system 140 includes a model storage unit 24, a correction model acquisition unit 50, and a learning unit 60.

[0105] The learning unit 60 acquires a teacher image in which an object to be inspected is captured, and generates a judgment model. The learning unit 60 includes a teacher image acquisition unit 61, a spectrum correction unit 62, and a judgment model generation unit 63.

[0106] The teacher image acquiring unit 61 acquires a teacher image 430 (see FIG. 13A). The teacher image 430 is an image obtained by capturing an image of a subject 413 (see FIG. 12) including a first material 218 and a background 219 by the imaging device 2. The teacher image acquiring unit 61 extracts one or more images 226 of the first material 218 and an image 227 of the background 219 by, for example, edge extraction. The teacher image acquiring unit 61 calculates a parameter indicating the size, a parameter indicating the shape, and a parameter indicating the arrangement of the first material 218 for each of the first materials 218 by a method similar to that of the reference image acquiring unit 11. The teacher image acquiring unit 61 acquires, for each of the first materials 218, a spectrum of a pixel included in the image 226 of the first material 218 in the teacher image 430 as a first spectrum 319. The teacher image acquiring unit 61 associates the first spectrum 319 for each of the first substances 218 with parameters indicating the size, shape, and arrangement of the first substances 218, and outputs the first spectrum 319 to the spectrum correcting unit 62. The teacher image acquiring unit 61 also outputs the spectrum of the pixels included in the image 227 of the background 219 in the teacher image 430 to the spectrum correcting unit 62 as the reflection spectrum 318 of the background (see FIG. 13B). The teacher image acquiring unit 61 also outputs the entire teacher image 430 to the spectrum correcting unit 62.

[0107] The spectrum correction unit 62 corrects the first spectrum 319 of the image 226 of the first material 218 in the teacher image 430, similarly to the spectrum correction unit 32 according to the embodiment. The spectrum correction unit 62 estimates the value of the background spectrum influence α for each of the images 226 of the first material 218, based on a combination of parameters indicating the size, shape, and arrangement of the first material 218, and a correction model. Through this process, the value of the background spectrum influence α in the teacher image 430 is calculated based on the estimation that if the size, shape, and arrangement of the subject are similar between the first reference image 421 and the teacher image 430, then a similar background spectrum influence α exists.

[0108] The spectrum correction unit 62 calculates the influence α of the background spectrum for all pixels included in the image 226 of any one of the first materials 218 in the teacher image 430. The spectrum correction unit 62, like the spectrum correction unit 32, calculates a second spectrum 320 for each pixel based on the background reflection spectrum 318, the influence α of the background spectrum for each pixel, and the first spectrum 319 of the pixel. The spectrum correction unit 62 outputs an image in which the first spectrum 319 is replaced with the second spectrum 320 for all pixels included in the image 226 of any one of the first materials 218 among the pixels included in the teacher image 430 to the determination model generation unit 63 as a corrected image.

[0109] The judgment model generating unit 63 generates teacher data in which the second spectrum 320 of the pixel included in the image 226 of the first material 218 in the corrected image is input data and information indicating the first material is a correct answer. The judgment model generating unit 63 performs machine learning using the above-mentioned teacher data to generate a judgment model. For example, a support vector machine or a random forest can be used as the machine learning. Note that the machine learning is not limited to this, and any machine learning using teacher data may be used, for example, a neural network. The judgment model generating unit 63 outputs the generated judgment model to the judgment model holding unit 22.

[0110] 2. Operation The following describes the operation of the model generation system 150. Fig. 14 is a flowchart showing the operation of the model generation system 150 according to Modification 2. Note that the same operations as those of the inspection system 100 shown in Fig. 9 are denoted by the same step numbers, and detailed descriptions thereof will be omitted.

[0111] The model generation system 150 acquires reference images (step S1). The reference image acquisition unit 11 of the model generation system 150 acquires a first reference image 421, a second reference image 422, and a third reference image. Details are the same as those in the first embodiment, and therefore will be omitted.

[0112] Next, the model generation system 150 extracts information from the reference images (step S2). The reference image acquisition unit 11 of the model generation system 150 extracts information required to generate a correction model from each of the first reference image 421, the second reference image 422, and the third reference image. Details are the same as those in the first embodiment, and therefore will be omitted.

[0113] Next, the model generation system 150 generates a correction model (step S3). The correction model generation unit 12 of the model generation system 150 generates a correction model based on information extracted from the reference image. Details are the same as those in the first embodiment, so they are omitted.

[0114] Next, the model generation system 150 acquires a teacher image 430 (step S8). The teacher image acquisition unit 61 of the model generation system 150 acquires, from the imaging device 2, a teacher image 430 obtained by capturing an image of a subject 413 including a first material 218 and a background 219.

[0115] Next, the model generation system 150 extracts information of the teacher image 430 (step S9). The teacher image acquisition unit 61 of the model generation system 150 extracts one or more images 226 of the first material 218 and an image 227 of the background 219 from the teacher image 430. The teacher image acquisition unit 61 calculates parameters indicating the size, shape, and arrangement of each of the first materials 218 based on the image 226. The teacher image acquisition unit 61 acquires, for each of the first materials 218, a spectrum of a pixel included in the image 226 of the first material 218 in the teacher image 430 as a first spectrum 319. The teacher image acquisition unit 61 outputs the first spectrum 319 to the spectrum correction unit 62 in association with the parameters indicating the size, shape, and arrangement of the first material 218 for each of the first materials 218. Furthermore, the teacher image acquiring unit 61 outputs the spectrum of the pixels included in the image 227 of the background 219 in the teacher image 430 to the spectrum correcting unit 62 as the background reflection spectrum 318. Furthermore, the teacher image acquiring unit 61 outputs the entire teacher image 430 to the spectrum correcting unit 62.

[0116] Next, the model generation system 150 corrects the spectrum (step S10). The spectrum correction unit 62 of the model generation system 150 estimates the value of the background spectrum influence α for each of the first materials 218 based on a combination of parameters indicating the size, shape, and arrangement of the first materials 218, and the correction model. The spectrum correction unit 62 calculates the second spectrum 320 for each pixel based on the reflection spectrum 318 of the background 219, the background spectrum influence α for each pixel, and the first spectrum 319 of the pixel. The spectrum correction unit 62 outputs an image in which the first spectrum 319 is replaced with the second spectrum 320 for all pixels included in the image 226 of any one of the first materials 218 among the pixels included in the teacher image 430, to the determination model generation unit 63 as a corrected image.

[0117] Next, the model generation system 150 creates a judgment model (step S11). The judgment model generation unit 63 of the model generation system 150 creates training data in which the second spectrum 320 of the pixel included in the image 226 of the first material 218 in the corrected image is input data and information indicating the first material is a correct answer. The judgment model generation unit 63 performs machine learning using the training data described above to generate a judgment model.

[0118] Through the above processing, a determination model is generated that uses the second spectrum 320 as an input and determines the first material.

[0119] 3.Effects The determination model according to the second modification is a machine learning model that determines the first substance based on the second spectrum 320. Therefore, by using the determination model according to the second modification in the inspection system 100 according to the embodiment or the determination system 110 according to the first modification, it is possible to improve the determination accuracy of the first substance. In particular, since the second spectrum 320 for generating the determination model and the second spectrum 317 in the determination system are corrected in the same manner, it is possible to reduce the influence of the background spectrum.

[0120] Also, similarly to the first embodiment, in the second spectrum 320 after correction by the spectrum correction unit 62, the influence of the background spectrum 318 is reduced to the same extent as the second reference spectrum, compared to the first spectrum 319 of each pixel of the teacher image 430. That is, the input in the teacher data of the judgment model is less influenced by the background spectrum 318. Therefore, the influence of the background spectrum in the judgment model can be reduced. That is, the accuracy of the judgment model can be improved.

[0121] (Other Modifications of the Embodiments) (1) In the embodiment and the modified example, the reference image acquisition unit 11 calculates the maximum diagonal distance as the dimension, the area ratio as the shape, and the angle between the maximum diagonal and the x-axis as the arrangement for the second material 213. The correction model generation unit 12 generates a correction model using the dimension, shape, and arrangement as input. The inspection image acquisition unit 31 calculates the dimension, shape, and arrangement of the inspection object 216 based on the image 224 of the inspection object 216, and the spectrum correction unit 32 corrects the spectrum using the dimension, shape, and arrangement of the inspection object 216.

[0122] However, the input to the correction model is not limited to the above example. The input to the correction model may be any element that affects the value of the background spectrum influence α on the first reference spectrum of the image 221 of the second material 213 in the first reference image 421 or the first spectrum of the image 224 of the inspection object 216 in the inspection object image 420. The element used as the input to the correction model may be one or more of the size, shape, and arrangement, or may be other elements.

[0123] (2) In the embodiment and each modification, the correction model generating unit 12 sets the spectrum 322 of the image 222 of the object 214 in the second reference image 422 as the reflection spectrum of the second material. In addition, the correction model generating unit 12 sets the spectrum 323 of the image 223 of the background 215 in the third reference image as the reflection spectrum of the background. However, the correction model generating unit 12 may use the reflection spectrum of the second material and the reflection spectrum of the background, which are known data. In this case, the reference image acquiring unit 11 may output the image of the object 214 in the second reference image 422 in combination with parameters indicating the size, shape, and arrangement of the object 214. In addition, the correction model generating unit 12 may calculate the influence α of the background spectrum for the image of the object 214 in the second reference image 422 as well, and use it as part of the teacher data for generating the correction model.

[0124] (summary) The inspection system (100; 130) according to the first aspect includes an inspection image acquisition unit (31), a spectrum correction unit (32), a spectrum determination unit (33), and a result output unit (40). The inspection image acquisition unit (31) acquires an inspection object image (420). The inspection object image (420) is obtained by capturing an image of a subject (410) including an inspection object (216) and a background (217) in four or more wavelength ranges. The spectrum correction unit (32) performs correction on a first spectrum (315) based on a spectrum (316) of an image (225) of the background (217) to generate a second spectrum (317). The first spectrum (315) is the spectrum of an image (224) of the inspection object (216) in the inspection object image (420). The spectrum determining section (33) determines whether the test object (216) is the first substance based on the second spectrum (317).

[0125] According to the inspection system (100; 130) of the first aspect, it is possible to reduce the influence of the spectrum (316) of the image (225) of the background (217) on the accuracy of determining whether the inspection object (216) is the first material, and therefore to improve the accuracy of identifying the inspection object (216).

[0126] In the inspection system (100; 130) according to the second aspect, in the first aspect, the spectrum correction unit (32) corrects the first spectrum (315) based on the first reference spectrum (321) and the second reference spectrum (322) to generate the second spectrum (317). The first reference spectrum (321) is the spectrum of the image (221) of the second material (213) in the first reference image (421). The second reference spectrum (322) is the spectrum of the image (222) of the second material (214) in the second reference image (422). The first reference image (421) is obtained by imaging the second material (213) including a portion having at least one of a size, a shape, and an arrangement that is known and a width that is less than a threshold value. A second reference image (422) is obtained by imaging a second material (214) having a known size, shape, and / or orientation and a width greater than or equal to a threshold value.

[0127] According to the inspection system (100; 130) of the second aspect, a correction is performed to reduce the influence of the spectrum (316) of the background (217) based on the first reference spectrum (321) and the second reference spectrum (322), which are two spectra of the same material but different dimensions. Therefore, the influence of the spectrum (316) of the background (217) can be reduced with high accuracy.

[0128] In the inspection system (100; 130) according to the third embodiment, the first substance and the second substance are different substances from those in the second embodiment.

[0129] According to the inspection system (100; 130) of the third aspect, the second material that is the subject of the first reference image (421) and the second reference image (422) does not need to be the same as the first material that may be included in the subject of the inspection target image (420). Therefore, it is not necessary to generate a correction model for each first material, and the generation of the correction model can be simplified.

[0130] In the inspection system (100; 130) according to the fourth aspect, in the second or third aspect, the first reference image (421) is an image obtained by capturing an image of a particle (213) of the second material.

[0131] According to the inspection system (100; 130) of the fourth aspect, the particulate second substance (213) can be imaged to generate the first reference image (421). Therefore, the correction model can be easily generated. In addition, since a spectrum correction model suitable for a particulate subject (410) is generated, when the inspection target (216) is particulate, the spectrum can be corrected with high accuracy, and the identification accuracy of the inspection target (216) can be improved.

[0132] In the inspection system (100; 130) according to the fifth embodiment, in the fourth embodiment, the second reference image (422) is an image obtained by capturing an image of particles (214) of the second material.

[0133] According to the inspection system (100; 130) of the fifth aspect, a particulate second substance can be imaged to generate a second reference image (422). Therefore, the first reference image (421) and the second reference image (422) can be easily prepared by imaging the second substance having a different particle size.

[0134] In the inspection system (100; 130) according to the sixth aspect, in the second or third aspect, the first reference image (421) is an image obtained by capturing an image of a TEG pattern (213) containing a second substance.

[0135] According to the inspection system (100; 130) of the sixth aspect, a spectrum correction model suitable for the subject (410) that is a TEG pattern is generated. Therefore, when the inspection target (216) is a TEG pattern, the spectrum can be corrected with high accuracy, and the identification accuracy of the inspection target (216) can be improved.

[0136] In the inspection system (100; 130) according to the seventh embodiment, in the sixth embodiment, the second reference image (422) is an image obtained by capturing an image of a TEG pattern (214) containing a second substance.

[0137] According to the inspection system (100; 130) of the seventh aspect, a sample is created using a second substance to have two types of TEG patterns with different widths, and a first reference image (421) and a second reference image (422) can be easily prepared by imaging the sample.

[0138] In the inspection system (100; 130) according to the eighth aspect, in any one of the second to seventh aspects, the result output section (40) outputs at least one of the first reference spectrum (321) and the second reference spectrum (322).

[0139] According to the inspection system (100; 130) of the eighth embodiment, the user can confirm one or both of the first reference spectrum (321) and the second reference spectrum (322). Therefore, the user can confirm whether the spectrum correction by the spectrum correction unit (32) is appropriate.

[0140] In the inspection system (100; 130) according to the ninth aspect, in any one of the first to eighth aspects, the result output section (40) outputs the second spectrum (317).

[0141] According to the inspection system (100; 130) of the ninth aspect, the user can check the second spectrum (317). Therefore, the user can check at least one of whether the spectrum correction by the spectrum correction unit (32) is appropriate or not and whether the judgment by the spectrum judgment unit (33) is appropriate or not.

[0142] In the inspection system (100; 130) according to the tenth aspect, in any one of the first to ninth aspects, the result output section (40) outputs the first spectrum (315).

[0143] According to the inspection system (100; 130) of the tenth aspect, the user can check the first spectrum (315). Therefore, the user can check whether the spectrum correction performed by the spectrum correction unit (32) is appropriate.

[0144] A model generation system (120) according to an eleventh aspect includes a reference image acquisition unit (11) and a correction model generation unit (12). The reference image acquisition unit (11) acquires a first reference image (421) and a second reference image (422). The first reference image (421) and the second reference image (422) are obtained by capturing images of a predetermined material, at least one of the dimensions, shape, and arrangement, and a background (215) in four or more wavelength ranges. The correction model generation unit (12) generates a spectrum correction model using teacher data based on the first reference spectrum (321) and the second reference spectrum (322). The first reference spectrum (321) is a spectrum of an image (221) of the predetermined material (213) in the first reference image (421). The second reference spectrum (322) is the spectrum of the image (222) of the predetermined material (214) in the second reference image (422). The first reference image (421) is obtained by imaging a predetermined material (213) having at least one of a size, a shape, and an arrangement known and including a portion having a width less than a threshold. The second reference image (422) is obtained by imaging a predetermined material (214) having at least one of a size, a shape, and an arrangement known and having a width equal to or greater than a threshold. The spectrum correction model is a machine learning model for correcting the spectrum (315) of the image (224) of the inspection object (216) in the inspection object image (420) based on the spectrum (316) of the image (225) of the background (217). The inspection object image (420) is obtained by imaging an object (410) including the inspection object (216) and the background (217) in four or more wavelength ranges.

[0145] According to the model generation system (120) of the eleventh aspect, a correction model is generated based on a first reference spectrum (321) and a second reference spectrum (322), which are two spectra made of the same material but different in size. By using this correction model, a correction can be performed to reduce the influence of the spectrum (316) of the image (225) of the background (217) on the spectrum (315) of the image (224) of the inspection object (216).

[0146] A determination system (110) according to a twelfth aspect includes a correction model acquisition unit (50), an inspection image acquisition unit (31), a spectrum correction unit (32), a spectrum determination unit (33), and a result output unit (40). The correction model acquisition unit (50) acquires a spectrum correction model from a model generation system (120) according to the eleventh aspect. The inspection image acquisition unit (31) acquires an inspection object image (420) obtained by capturing an image of a subject (410) including an inspection object (216) and a background (217) in four or more wavelength ranges. The spectrum correction unit (32) corrects a first spectrum (315), which is a spectrum of an image (224) of the inspection object (216) in the inspection object image (420), using the spectrum correction model to generate a second spectrum (317). The spectrum determining section (33) determines whether or not the inspection object (216) is the first substance based on the second spectrum (317). The result output section (40) outputs the result of the determination by the spectrum determining section (33).

[0147] According to the determination system (110) of the twelfth aspect, it is possible to reduce the influence of the spectrum (316) of the image (225) of the background (217) on the accuracy of determining whether the test object (216) is the first material or not, and therefore to improve the accuracy of identifying the test object (216).

[0148] The inspection method according to the thirteenth aspect acquires an inspection object image (420), corrects a first spectrum (315) that is the spectrum of an image (224) of an inspection object (216) in the inspection object image (420) based on a spectrum (316) of an image (225) of a background (217) to generate a second spectrum (317), determines whether the inspection object (216) is a predetermined material based on the second spectrum (317), and outputs the determination result. The inspection object image (420) is obtained by capturing an image of a subject (410) including the inspection object (216) and the background (217) in four or more wavelength bands.

[0149] According to the inspection method of the thirteenth aspect, it is possible to reduce the influence of the spectrum (316) of the image (225) of the background (217) on the accuracy of determining whether the inspection object (216) is the first material or not, and therefore to improve the accuracy of identifying the inspection object (216).

[0150] A program according to a fourteenth aspect is a program that causes one or more processors to execute the inspection method according to the thirteenth aspect.

[0151] According to the program of the fourteenth aspect, a processor executing the program functions as an inspection system (100; 130). Therefore, the inspection method executed by the processor can reduce the influence of the spectrum (316) of the image (225) of the background (217) on the accuracy of determining whether the inspection object (216) is the first material or not. Therefore, the accuracy of identifying the inspection object (216) can be improved.

[0152] A model generation method according to a fifteenth aspect includes acquiring a first reference image (421) and a second reference image (422), and generating a spectrum correction model using training data based on the first reference spectrum (321) and the second reference spectrum (322). The first reference image (421) and the second reference image (422) are obtained by capturing images of a predetermined material, at least one of the dimensions, shape, and arrangement, and a background (215) in four or more wavelength ranges. The first reference spectrum (321) is the spectrum of an image (221) of the predetermined material (213) in the first reference image (421). The second reference spectrum (322) is the spectrum of an image (222) of the predetermined material (214) in the second reference image (422). The first reference image (421) is obtained by imaging a predetermined material (213) having at least one of known dimensions, shape, and arrangement and including a portion whose width is less than a threshold. The second reference image (422) is obtained by imaging a predetermined material (214) having at least one of known dimensions, shape, and arrangement and including a portion whose width is equal to or greater than a threshold. The spectrum correction model is a machine learning model for correcting the spectrum (315) of the image (224) of the inspection object (216) based on the spectrum (316) of the image (225) of the background (217) in the inspection object image (420). The inspection object image (420) is obtained by imaging an object (410) including the inspection object (216) and the background (217) in four or more wavelength bands.

[0153] According to the model generating method of the fifteenth aspect, a correction model is generated based on a first reference spectrum (321) and a second reference spectrum (322), which are two spectra made of the same material but different in size. By using this correction model, a correction can be performed to reduce the influence of the spectrum (316) of the image (225) of the background (217) on the spectrum (315) of the image (224) of the inspection object (216).

[0154] A program according to a sixteenth aspect is a program that causes one or more processors to execute the model generation method according to the fifteenth aspect.

[0155] According to the program of the sixteenth aspect, a processor executing the program functions as a model generation system (120). Thus, a model generation method executed by the processor generates a correction model based on a first reference spectrum (321) and a second reference spectrum (322), which are two spectra having the same material but different dimensions. By using this correction model, a correction can be performed to reduce the influence of the spectrum (316) of the image (225) of the background (217) on the spectrum (315) of the image (224) of the inspection object (216). [Explanation of symbols]

[0156] 100, 130 Inspection System 110 Judgment System 120 Model Generation System 11 Reference image acquisition section 12 Correction model generation unit 110 Judgment System 31 Inspection image acquisition unit 32 Spectral correction section 33 Spectral evaluation unit 40 Result output section 50 Correction model acquisition unit 213, 214 Second substance (prescribed substance) 216 Test Subjects 215, 217 background 221, 222 Images of the second substance 224 Image of the subject of examination 223, 225 Statue in the background 315 The First Spectrum 316 Background Image Spectrum 317 The Second Spectrum 321 First Reference Spectrum 322 Second Reference Spectrum 410 Subject 420 Inspection target image 421 First reference image 422 Second Reference Image

Claims

1. an inspection image acquisition unit that acquires an inspection object image obtained by capturing an image of a subject including an inspection object and a background in four or more wavelength ranges; a spectrum correction unit that performs correction on a first spectrum, which is a spectrum of the image of the inspection object in the inspection object image, based on a spectrum of the image of the background to generate a second spectrum; a spectrum determination unit that determines whether the inspection object is a first substance based on the second spectrum; a result output unit that outputs a result of the determination by the spectrum determination unit; the spectrum correction unit performs the correction on the first spectrum based on a first reference spectrum and a second reference spectrum to generate the second spectrum; the first reference spectrum is a spectrum of an image of a second material in the first reference image; the second reference spectrum is a spectrum of an image of the second material in a second reference image; the first reference image is obtained by imaging the second material including a portion having at least one of a known size, shape, and arrangement and a width less than a threshold; The second reference image is obtained by imaging the second material having at least one of a size, a shape, and a configuration that is known and a width that is equal to or greater than the threshold value. Inspection system.

2. The first substance and the second substance are different substances. The inspection system of claim 1 .

3. The first reference image is an image obtained by imaging particles of the second material.

3. The inspection system according to claim 1 or 2.

4. The second reference image is an image obtained by imaging particles of the second material. The inspection system of claim 3 .

5. The first reference image is an image obtained by imaging a TEG pattern including the second substance.

3. The inspection system according to claim 1 or 2.

6. The second reference image is an image obtained by imaging a TEG pattern including the second substance. The inspection system of claim 5 .

7. The result output unit outputs at least one of the first reference spectrum and the second reference spectrum. The inspection system according to any one of claims 1 to 6.

8. The result output unit outputs the second spectrum. The inspection system according to any one of claims 1 to 7.

9. The result output unit outputs the first spectrum. The inspection system according to any one of claims 1 to 8.

10. A reference image acquisition unit that acquires a first reference image and a second reference image obtained by capturing an image of a predetermined material, at least one of which has known dimensions, shape, and arrangement, and a background in four or more wavelength ranges; a correction model generating unit that generates a spectrum correction model by using teacher data based on a first reference spectrum that is a spectrum of an image of the predetermined material in the first reference image and a second reference spectrum that is a spectrum of an image of the predetermined material in the second reference image, The first reference image is obtained by imaging the predetermined material having at least one of a size, a shape, and a configuration that is known and a portion having a width less than a threshold value; The second reference image is obtained by imaging the predetermined material having at least one of a size, a shape, and a position known and a width equal to or greater than the threshold value; The spectrum correction model is a machine learning model for performing correction on the spectrum of an image of an inspection object based on the spectrum of an image of the background in an inspection object image obtained by capturing an image of an object including the inspection object and a background in four or more wavelength ranges. Model generation system.

11. A correction model acquisition unit that acquires a spectral correction model from the model generation system according to claim 10; an inspection image acquisition unit that acquires an inspection object image obtained by capturing an image of a subject including an inspection object and a background in four or more wavelength ranges; a spectrum correction unit that corrects a first spectrum, which is a spectrum of the image of the inspection object in the inspection object image, by using the spectrum correction model to generate a second spectrum; a spectrum determination unit that determines whether the inspection object is a first substance based on the second spectrum; A result output unit that outputs a result of the judgment by the spectrum judgment unit. Judging system.

12. An image of an object to be inspected is obtained by imaging a subject including the object to be inspected and a background in four or more wavelength ranges; generating a second spectrum by performing a correction on a first spectrum, which is a spectrum of the image of the inspection object in the inspection object image, based on a spectrum of the image of the background; determining whether the test object is a predetermined substance based on the second spectrum; A testing method comprising: When generating the second spectrum, the correction is performed on the first spectrum based on a first reference spectrum and a second reference spectrum to generate the second spectrum; the first reference spectrum is a spectrum of an image of a second material in the first reference image; the second reference spectrum is a spectrum of an image of the second material in a second reference image; the first reference image is obtained by imaging the second material including a portion having at least one of a known size, shape, and arrangement and a width less than a threshold; The second reference image is obtained by imaging the second material having at least one of a size, a shape, and a configuration that is known and a width that is equal to or greater than the threshold value. Testing method.

13. A method for testing a computer-implemented computer-implemented system comprising: program.

14. A method for obtaining a first reference image and a second reference image obtained by imaging a predetermined material, at least one of which is known in terms of size, shape, and arrangement, and a background in four or more wavelength ranges; A model generation method for generating a spectrum correction model using training data based on a first reference spectrum that is a spectrum of an image of the predetermined material in the first reference image and a second reference spectrum that is a spectrum of an image of the predetermined material in the second reference image, the model generation method comprising: The first reference image is obtained by imaging the predetermined material having at least one of a size, a shape, and a configuration that is known and a portion having a width less than a threshold value; the second reference image is obtained by imaging the predetermined material including a portion having a known size, shape, and arrangement and a width equal to or greater than the threshold value; The spectrum correction model is a machine learning model for performing correction on the spectrum of an image of an inspection object based on the spectrum of an image of the background in an inspection object image obtained by capturing an image of an object including the inspection object and a background in four or more wavelength ranges. Model generation method.

15. A method for generating a model according to claim 14, comprising the steps of: program.

Images