Image quality learning device, image quality judgment device and image reading device

Abstract

The image quality learning device (2) of the present invention learns the judgment result of whether a planar object (1) is a good product or a defective product from at least one of the three-dimensional shape or color of the surface of the planar object (1), including: a surface image input unit (3), which inputs two-dimensional data as image data of the surface of the planar object (1); a judgment information input unit (4), which inputs judgment information indicating that the planar object (1) corresponding to the two-dimensional data is either a good product or a defective product; and a learning unit (5), which learns the existence area (1R) of the three-dimensional shape or color of the surface in the two-dimensional data as the basis of the judgment information based on the two-dimensional data and the judgment information.

Description

Technical Field

[0001] The present disclosure relates to an image quality learning device for learning a determination result of whether a planar object to be inspected is a good product or a defective product, an image quality determination device using the device, and an image reading device. Background Art

[0002] In the past, there were image reading devices for inspecting the presence of defects such as damage in planar objects (sheet objects) such as films that are inspection objects (for example, see Patent Documents 1 and 2). Planar objects include, in addition to films, printed materials, foils, cloths, panels (plates), labels (printed labels), semiconductor wafers, substrates (masks), and the like. The image reading devices disclosed in Patent Documents 1 and 2 use both reflected light and transmitted light from the film in the inspection in order to inspect films that transmit visible light. However, in image reading devices for inspecting planar objects, at least one of reflected light or transmitted light can be used as needed. In order to inspect planar objects, invisible light such as infrared or ultraviolet rays can also be used instead of visible light. Both visible light and invisible light can be used for inspecting planar objects. In addition, in image reading devices for inspecting planar objects, a learning model obtained by machine learning using AI (Artificial Intelligence) and the like is used (for example, see Patent Documents 3 and 4).

[0003] Patent Document 3 discloses a method for identifying the type of defect in an inspection object based on data accumulated from machine learning results related to identifying defect types contained in line-segmented images with different brightness and appearance, even when images of the same inspection object are captured. Patent Document 4 discloses a method for comparing AI-generated benchmark data with images of the inspection object, and using AI to determine whether the inspection object is good or bad based on the matching ratio obtained from the comparison results.

[0004] Furthermore, in an image reading device for inspecting an inspection object, the quality of the physical label itself is not determined, but rather the pass / fail status of the inspection is determined based on information on the label that a person has assigned to the inspection object (see, for example, Patent Document 5). Patent Document 5 discloses a learning unit that uses image data to perform machine learning on the relationship between the image data and the pass / fail status of the inspection object.

[0005] Image reading devices for inspecting inspection objects include those using line sensors such as erecting equal-magnification optical systems (see, for example, Patent Documents 1, 2, and 3) and those using area sensors such as reduction optical systems or cameras (see, for example, Patent Documents 4 and 5). Furthermore, image reading devices for inspecting inspection objects include those with a built-in light source for irradiating light onto the inspection object and those with an external light source.

[0006] Prior art literature

[0007] Patent Literature

[0008] Patent Document 1: Japanese Patent Application Laid-Open No. 2015-68670

[0009] Patent Document 2: Japanese Patent Application Laid-Open No. 2019-23587

[0010] Patent Document 3: Japanese Patent Application Laid-Open No. 2019-23588

[0011] Patent Document 4: Japanese Patent Application Laid-Open No. 2019-56591

[0012] Patent Document 5: Japanese Patent Application Laid-Open No. 2019-184305 Summary of the Invention

[0013] Technical problem to be solved by the invention

[0014] However, there is a problem that conventional machine learning is a learning model whose basis for determining information is unclear.

[0015] The purpose of the present disclosure is to solve the above-mentioned problems, and its purpose is to obtain an image quality learning device that learns the judgment result of whether a planar object is a good or bad product based on at least one of the three-dimensional shape or color of the surface of the planar object, an image quality judgment device using the device, and an image reading device.

[0016] Technical means for solving technical problems

[0017] The image quality learning device involved in the present disclosure learns the judgment result of whether the planar object is a good product or a defective product from at least one of the three-dimensional shape or color of the surface of the planar object, and is characterized in that it includes: a surface image input unit, which inputs two-dimensional data as image data of the surface of the planar object; a judgment information input unit, which inputs the judgment result, i.e., judgment information, indicating whether the planar object corresponding to the two-dimensional data is a good product or a defective product; and a learning unit, which learns the existence area of the three-dimensional shape or the color of the surface in the two-dimensional data as the basis of the judgment information based on the two-dimensional data and the judgment information.

[0018] The image quality judgment device involved in the present disclosure uses the learning results of an image quality learning device, which learns the judgment result of whether the planar object is a good product or a defective product from at least one of the three-dimensional shape or color of the surface of the planar object. The image quality learning device includes: a surface image input unit, which inputs two-dimensional data as image data of the surface of the planar object; a judgment information input unit, which inputs the judgment result, i.e., judgment information, indicating whether the planar object corresponding to the two-dimensional data is a good product or a defective product; and a learning unit, which learns the existence area of the three-dimensional shape or the color of the surface in the two-dimensional data as the basis of the judgment information based on the two-dimensional data and the judgment information, and is characterized in that the image quality learning device includes: a new surface image input unit, which inputs new two-dimensional data obtained after newly reading the planar object; and an image quality judgment unit, which judges whether the planar object corresponding to the new two-dimensional data is a good product or a defective product based on the learning result learned by the learning unit.

[0019] The image reading device involved in the present disclosure includes an image quality judgment device, which uses the learning results of an image quality learning device, and the image quality learning device learns the judgment result of whether the planar object is a good product or a defective product from at least one of the three-dimensional shape or color of the surface of the planar object. The image quality learning device includes: a surface image input unit, which inputs two-dimensional data as image data of the surface of the planar object; a judgment information input unit, which inputs the judgment result, i.e., judgment information, indicating whether the planar object corresponding to the two-dimensional data is a good product or a defective product; and a learning unit, which learns the judgment result based on the two-dimensional data. and the judgment information to learn the three-dimensional shape of the surface or the existence area of the color in the two-dimensional data which is the basis of the judgment information, the image quality judgment device includes: a new surface image input unit, which inputs the new two-dimensional data obtained after the planar object is newly read; and an image quality judgment unit, which determines whether the planar object corresponding to the new two-dimensional data is a good product or a defective product based on the learning result learned by the learning unit, and is characterized in that the image reading device includes: an optical unit for converging light from the planar object, and a sensor unit for receiving the light converged by the optical unit and generating the new two-dimensional data.

[0020] Effects of the Invention

[0021] According to the present disclosure, the image quality learning device obtained can obtain a learning result (learning model) of the three-dimensional shape or color existence area of the surface of a planar object, thereby obtaining an image quality judgment device and an image reading device with a clear position as the basis for judgment information in the image data of the surface of a planar object. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 This is a functional block diagram of the image quality learning device according to the first embodiment.

[0023] Figure 2 This is a flowchart illustrating the operation of the image quality learning device (image quality learning method) according to the first embodiment.

[0024] Figure 3 1 and 2 are diagrams illustrating examples of image data input to the image quality learning device according to the first embodiment (new image data input to the image reading device according to the first embodiment).

[0025] Figure 4 This is a functional block diagram of the image quality determination device and the image reading device according to the first embodiment.

[0026] Figure 5 This is a flowchart illustrating the operation (image quality determination method) of the image quality learning device, the image quality determination device, and the image reading device according to the first embodiment.

[0027] Figure 6 This is a functional block diagram of the image reading device according to the first embodiment. DETAILED DESCRIPTION

[0028] Implementation method 1.

[0029] Below, use Figures 1 to 6 The image quality learning device involved in embodiment 1, the image quality determination device using the device, and the image reading device (the image quality determination device involved in embodiment 1, the image reading device involved in embodiment 1) are described. In the figures, the same reference numerals represent the same or corresponding parts, and their detailed description is omitted. Figure 3 In this application, the planar object 1 to be inspected is, for example, a sheet of printed matter, film, foil, cloth, panel (board), label (printed label), semiconductor wafer, substrate (mask), or the like. Learning in this application assumes the presence of multiple planar objects 1. Therefore, as the amount of learning increases, even when a newly captured image, described later, is an image of an unknown planar object 1, it is possible to determine whether the image is a good or bad product.

[0030] exist Figure 1 and Figure 3 In the embodiment, the image quality learning device 2 learns the judgment result of whether the planar object 1 is a good product or a bad product from at least one of the three-dimensional shape or color of the surface of the planar object 1 (for example, from the image of one side), and constructs a learning model (the image quality learning device involved in embodiment 1). The three-dimensional shape of the surface of the planar object 1 refers to the three-dimensional shape of the planar object 1 set on the surface of the planar object 1 in the thickness direction. For example, it is a shape that changes in the thickness direction of the planar object 1 set on the surface of the planar object 1, or a concave and convex shape in the thickness direction of the planar object 1 set on the surface of the planar object 1. In addition, the surface color of the planar object 1 refers to, for example, a change in the surface color of the planar object 1. It can be considered that color is represented by three attributes: hue, brightness, and saturation.

[0031] exist Figure 1 and Figure 3 In the embodiment, the surface image input unit 3 inputs image data of the surface of the planar object 1, that is, two-dimensional data. The input image data is, for example, image data obtained by photographing the surface of the planar object 1. Figure 1 (A) and Figure 3 In the example, the judgment information input unit 4 inputs judgment information indicating whether the planar object 1 corresponding to the two-dimensional data is a good or defective product. The learning unit 5 learns the three-dimensional shape or color area 1R of the surface in the two-dimensional data, which is the basis for the judgment information, based on the two-dimensional data and the judgment information.

[0032] Preferably, the surface image input unit 3 receives two-dimensional data (image data) representing at least one of the following: a pattern formed on the surface, concave-convex surfaces, or components mounted on the surface as a three-dimensional shape, or at least one of the following: a surface pattern (color pattern), transparency, or a printed wiring pattern as a color. The two-dimensional data (image data) is obtained by photographing at least one of the following: a pattern formed on the surface, concave-convex surfaces, or components mounted on the surface as a three-dimensional shape, or by photographing at least one of the following: a surface pattern (color pattern), transparency, or a printed wiring pattern as a color. The three-dimensional pattern refers to embroidery or dyeing patterns woven into fabric, such as cloth, through variations in the use of thread or the weaving method used to construct the fabric. The three-dimensional pattern and concave-convex surfaces can be considered to represent the smoothness of the surface. Furthermore, the pattern of the printed material, i.e., the pattern, as a color, corresponds to the surface pattern (color pattern).

[0033] The concave and convex portions formed on the surface refer to the concave and convex portions on the surface of printed matter, films, foils, cloths, panels (boards), labels (printed labels), semiconductor wafers, substrates (masks), etc., including depressions and through-holes. The components mounted on the surface are particularly equivalent to the convex portions formed in the concave and convex portions on the surface. Specifically, it refers to the items attached to printed matter, films, foils, cloths, and the components placed on panels (boards), labels (printed labels), semiconductor wafers, substrates (masks), etc. As a color, the pattern on the surface refers to the pattern formed on the surface of printed matter, films, foils, cloths, panels (boards), labels (printed labels), semiconductor wafers, substrates (masks), etc. These patterns can be said to be color patterns (including monochrome). The color patterns include test patterns (including monochrome) used for reading tests of image reading devices such as one-dimensional line sensors or cameras (for example, the image reading device 10 described later). Transparency refers to the transparency (based on visibility or invisibility) of printed matter, films, foils, cloths, panels (boards), labels (printed labels), etc. The printed wiring pattern refers to a printed wiring pattern formed on the surface of a printed material, a film, a panel (board), a semiconductor wafer, a substrate (mask), or the like.

[0034] Machine learning such as AI can be applied to the learning unit 5 (image quality learning device 2). The learning unit 5 (image quality learning device 2) constructs and accumulates a learning model. As the amount of learning increases, the learning unit 5 improves the accuracy of determining the area 1R where the three-dimensional shape or color of the surface in the two-dimensional data, which is the basis for the judgment information, exists based on the two-dimensional data and the judgment information. That is, at the beginning of learning, the surface portion in the two-dimensional data where the three-dimensional shape or color used to determine the quality exists is a relatively wide area that includes parts that have no direct relationship with the quality. On the other hand, if the learning progresses, the surface portion in the two-dimensional data where the three-dimensional shape or color used to determine the quality exists is an area that does not contain any parts that have no direct relationship with the quality judgment, or an area that almost does not contain any parts that have no direct relationship with the quality judgment. For example, when the three-dimensional shape of the surface is a component mounted on the surface, the former is the component (area) itself, and the latter is the component and its surrounding (area).

[0035] Thus, as the amount of learning increases, not only can the existing region 1R be determined (narrowed), but the image quality learning device 2 can also learn the existing region 1R from the beginning. Figure 1 As shown in (B), the information of the existence area 1R can be included in the two-dimensional data and input to the surface image input unit 3. Although omitted in the figure, the information of the existence area 1R can be input to the image quality learning device 2 by other means. For example, the information of the existence area 1R can be included in the judgment information and input to the judgment information input unit 4. Of course, a dedicated existence area information input unit can be set in the image quality learning device 2, in which the information of the existence area 1R is input to the learning unit 5. That is, in Figure 1 In the image quality learning device 2 shown in (A), the input two-dimensional data and the judgment information are associated in the learning unit 5. Figure 1 In the image quality learning device 2 shown in (B), the input two-dimensional data, the judgment information, and the information of the existence area 1R are associated in the learning unit 5. Figure 1 (A) The image quality learning device 2 described above corresponds to Figure 4 The image quality learning device 2 is described, but the Figure 1 (B) The image quality learning device 2 described in the embodiment corresponds to Figure 4 If the information on the existing region 1R is information that indicates the position on the planar object 1 , the information on the existing region 1R may be information on coordinates or a distance from an end of the planar object 1 .

[0036] Next, use Figure 2 The operation of the image quality learning device according to the first embodiment (the image quality learning method according to the first embodiment) will be described. Figure 2In the process, step 1 is a processing step of inputting two-dimensional data, which is image data of the surface of a planar object 1, into the surface image input unit 3. Step 2 is a processing step of inputting judgment information, which is a judgment result of determining whether the planar object 1 corresponding to the two-dimensional data is a good product or a defective product, into the judgment information input unit 4. The processing order of step 1 and step 2 is irrelevant. Step 1 and step 2 can also be processed simultaneously. Step 3 is a processing step of causing the learning unit 5 to learn the existence area 1R of the three-dimensional shape or color of the surface in the two-dimensional data that serves as the basis for the judgment information based on the two-dimensional data and the judgment information. Step 3 can also use the information of the above-mentioned existence area 1R.

[0037] The surface image input unit 3 can input two-dimensional data, which is image data composed of a plurality of linear one-dimensional data (short strips of image data, a portion (one column) of the image data of the planar object 1) arranged in an arrangement. For example, consider inputting two-dimensional data into the surface image input unit, which is composed of one-dimensional data sequentially acquired in a sub-scanning direction intersecting the main scanning direction by a one-dimensional line sensor (equivalent to an example of the image reading device 10 described later), and the one-dimensional line sensor reads the reading object (planar object 1) in the main scanning direction with the expansion direction of the linear one-dimensional data as the main scanning direction. In these cases, the learning unit 5 can separate and learn the location of the existence area 1R for each one-dimensional data. In addition, the learning unit 5 can also newly generate judgment information based on the existence area 1R of each one-dimensional data. In addition, it is also possible to newly generate judgment information based on the existence area 1R of each of the plurality of short strips of image data equivalent to a portion of the final image data (two-dimensional data). The plurality of short strips of image data can be continuous images or intermittent images.

[0038] In this application, what is referred to as one-dimensional data and new one-dimensional data is not virtual data that has a reading width only in the main scanning direction. In addition to the reading width in the main scanning direction, for convenience, the short strip of image data with a reading width in the sub-scanning direction of one pixel (sensor element) acquired in the image reading device 10 described later is referred to as one-dimensional data. Therefore, the reading width in the sub-scanning direction is affected by one pixel (sensor element). That is, a one-dimensional line sensor has a reading width in the sub-scanning direction of one pixel (sensor element) in addition to the reading width in the main scanning direction, but for convenience, it is referred to as a one-dimensional line sensor. In addition, the short strip of image data can also be input as two-dimensional data into the surface image input unit 3, and the learning unit 5 can learn good or bad products separately through the short strip of image data. Therefore, in this application, the short strip of image data is both one-dimensional data (new one-dimensional data) and two-dimensional data (new two-dimensional data). In other words, in this application, as long as the one-dimensional data (new one-dimensional data) is short strip of image data, it can be said to be two-dimensional data (new two-dimensional data). Furthermore, as described above, the short strip of image data may be a portion (one row) of the image data of the planar object 1. Furthermore, the plurality of short strips of image data described above can naturally be considered as two-dimensional data.

[0039] exist Figure 3 In the example, there is a region 1R surrounded by a dotted line. Figure 3 Although one existing region 1R is shown, a plurality of existing regions 1R may be located on the planar object 1 . Figure 3 (A) shows the position of the existing region 1R on the planar object 1 . Figure 3 (B) illustrates the position of the existing region 1R in image data formed by arranging a plurality of linear one-dimensional data. Figure 3 (C) shows the Figure 3 In the case of (B), an existing region 1R is located in a one-dimensional data. Figure 3 In (B), the region 1R exists in four one-dimensional data. Figure 3 (B) Figure 3 In the case of (C), as described above, the learning unit 5 can easily learn the location of the existence region 1R for each one-dimensional data item, and can also easily generate new determination information based on the existence region 1R of each one-dimensional data item. The existence region 1R in the new two-dimensional data (new surface image) described below also has the same relationship with the new planar object 1 from which the new two-dimensional data (new surface image) is obtained. Figure 3 (B) and Figure 3(C) shows an example in which the number of columns is small to make the sub-scanning range of the planar object 1 and the one-dimensional line sensor easier to understand. Although this depends on the relative size difference between the planar object 1 and the one-dimensional line sensor, in reality, the number of columns often increases.

[0040] The good or bad products in the judgment information are as follows, for example. For the image data of the printed matter, the good product has the desired color of the printed matter, the configuration / orientation / size of the printing result, and the configuration / orientation / size of the reference mark. On the other hand, the bad product has print misalignment / off-printing / white spots / color irregularities / damage of the printed matter. For the image data of the film, the good product has the desired color of the film, the smoothness of the film surface, and the transparency of the film. On the other hand, the bad product has damage / cracks / color irregularities / holes on the film. For the image data of the foil, the good product has the desired color of the foil and the smoothness of the foil surface. On the other hand, the bad product has damage / cracks / color irregularities / holes on the film. For the image data of the cloth, the good product has the desired color of the cloth, the direction / size of the grid, and the smoothness of the cloth surface. On the other hand, the bad product has color irregularities / shedding / fuzzing of the cloth. For the image data of the panel (plate), the good product has the desired color of the panel, the smoothness of the panel surface, and the size / direction / size of the shapes (components) on the panel surface. On the other hand, defective products have color irregularities / damages / cracks / holes on the panel.

[0041] Next, for the image data of the label (printed label), the qualified product has the desired color of the label, the smoothness of the label surface, and the width / direction / size of the label printing (one-dimensional code, two-dimensional code, line, text, etc.). On the other hand, the defective product has the printing misalignment / off-printing / white spots / color irregularities / damage of the label. For the image data of the semiconductor chip, the qualified product has the desired color of the semiconductor chip, the smoothness of the semiconductor chip surface, and the size / direction / size of the shapes (components) on the semiconductor chip surface. On the other hand, the defective product has the color irregularities / damage / cracks / holes on the semiconductor chip. For the image data of the substrate (mask), the qualified product has the color of the substrate, the smoothness of the substrate surface, the position / size of the holes on the substrate, the configuration / direction / size of the printed characters (one-dimensional code, two-dimensional code, line, text, etc.) on the substrate, the state of the solder on the substrate, the solder chamfer on the substrate, and the presence / configuration / direction / size of the mounted parts (components) on the substrate surface. On the other hand, substrates with irregular colors, damage, cracks, holes, or misaligned / falling / cracked / white spots / scratches on the printing are considered defective products.

[0042] So far, it has been explained that the judgment information may include information about the existing area and input it into the judgment information input unit 4. Alternatively, the image quality learning device 2 is provided with a dedicated existing area information input unit for inputting the existing area information into the learning unit 5. In this case, information about the position of image data serving as a basis for quality determination in image data of the planar object 1 to be inspected (image data obtained by photographing the planar object 1 to be inspected), such as the aforementioned printed matter image data, film image data, foil image data, cloth image data, panel (board) image data, label (printed label) image data, semiconductor wafer image data, substrate (mask) image data, etc., is used as the existing area information.

[0043] In this way, the learning unit 5 receives information about the existence area and learns it, thereby enabling the rapid construction of a learning model. Of course, the learning unit 5 learns the existence area of the three-dimensional shape or color of the surface of the planar object 1 in the two-dimensional data (image data) serving as the basis for the judgment information based on the two-dimensional data (image data) and the judgment information. Therefore, when a certain amount of two-dimensional data (image data) and judgment information are input into the learning unit 5, the existence area can be obtained by comparing the two-dimensional data (image data) with each other. For example, when multiple image data with the same composition and the same inspection object differ in quality, the area for judging quality, i.e., the existence area, can be determined based on the differences in the image data. In other words, when the two-dimensional data (image data) of the same planar object 1 has different judgment results (quality) for each piece of the two-dimensional data (image data) as a good or defective product based on the two-dimensional data (image data) and the judgment information, the learning unit 5 learns the existence area of the three-dimensional shape or color of the surface of the two-dimensional data serving as the basis for the judgment information based on the differences in these two-dimensional data (image data).

[0044] exist Figure 4 In the image quality determination device 6, the image quality determination device 6 uses Figure 1 and Figure 4 The learning result (learning model) of the image quality learning device 2 (image quality determination device of embodiment 1) is shown. New two-dimensional data (new surface image) obtained by newly reading the planar object 1 is input to the new surface image input unit 7. Here, to distinguish it from the two-dimensional data (surface image) used by the learning unit 5 (image quality learning device 2) to construct the learning model, the new two-dimensional data (new surface image) is simply referred to as "new" and includes the existing two-dimensional data (surface image). Therefore, the new two-dimensional data (new surface image) can be referred to as image data for determination.

[0045] exist Figure 4In the example, the image quality determination unit 8 determines whether the planar object 1 corresponding to the new two-dimensional data is a good product or a defective product based on the learning results learned by the learning unit 5. In addition, the image quality determination unit 8 can determine whether the planar object 1 corresponding to the new two-dimensional data is a good product or a defective product, and extract the basis area 9 corresponding to the existing area 1R in the new two-dimensional data as the basis for this determination. In this way, if the basis area 9 (existence area 1R) in the new two-dimensional data is determined, then if the image quality determination unit 8 determines that the product is defective, the defective part can be known. Of course, if the basis area 9 (existence area 1R) in the new two-dimensional data is determined, then if the image quality determination unit 8 determines that the product is good, it can be known that there is no problem with the basis area 9 (existence area 1R).

[0046] Likewise, in Figure 4 In the embodiment, the image reading device 10 is provided with an image quality determination device 6 (the image reading device of embodiment 1). The image reading device 10 includes an optical portion 11 and a sensor portion 12. Preferably, the image reading device 10 may further include an output portion 13. The optical portion 11 converges light (reflected light or transmitted light) from the planar object 1. The sensor portion 12 is a color sensor, and the optical portion 11 receives the converged light to generate new two-dimensional data. The output portion 13 outputs (transmits) the new two-dimensional data to the new surface image input portion 7 as new two-dimensional data (new surface image) obtained after the planar object 1 is newly read. The output portion 13 can be omitted, and the new two-dimensional data can be sent directly from the sensor portion 12 to the new surface image input portion 7. In this case, it can be said that the function of the output portion 13 is embedded in the sensor portion 12.

[0047] Next, use Figure 5 The operation of the image quality learning device (image reading device) according to the first embodiment (the image quality determination method according to the first embodiment) will be mainly described. Figure 5 In the example, step 11 is a processing step of inputting the new two-dimensional data (new surface image) obtained after the new reading of the planar object 1 into the new surface image input unit 7. Step 12 is a processing step of inputting the newly captured image from the new surface image input unit 7 to the learning unit 5 and using the learning model. Step 13 is a processing step of the image quality determination unit 8 determining whether the planar object 1 corresponding to the new two-dimensional data is a good product or a defective product based on the learning result (learning model) learned by the learning unit 5. In step 13, the image quality determination unit 8 can extract a basis area 9 corresponding to the existence area 1R in the new two-dimensional data as a basis for determination.

[0048] use Figure 6 , a preferred example of the optical section 11 of the image reading device 10 is described. Figure 6In the image scanning apparatus 10, the light source 14 is an illumination device such as an LED (Light Emitting Diode), an organic EL (Electro-Luminescence), or a discharge lamp, and is preferably a linear light source 14 extending in the main scanning direction. The light source 14 irradiates the planar object 1 with light, and the optical unit 11 converges the reflected or transmitted light. The light source 14 can be built into the image scanning apparatus 10, located external to the image scanning apparatus 10, or controlled by the image scanning apparatus 10. Of course, the image scanning apparatus 10 and the light source 14 can be controlled separately by a higher-level control device (not shown).

[0049] exist Figure 6 In (a), the camera 11a collects light (reflected light or transmitted light) from the planar object 1. The camera 11a and the sensor unit 12 can be called an area sensor. Figure 6 In (b), the lens array 11b is arranged with a plurality of upright equal-magnification optical system lenses. The sensor unit 12 is a sensor element array 12 in which a plurality of sensor elements corresponding to each upright equal-magnification optical system lens are arranged. The sensor element array 12 can output each new one-dimensional data (short strip image data, a part (one column) of the image data of the planar object 1) to the new surface image input unit 7. Thus, the image quality judgment device 6 can perform the following actions. First, in the structure of the basic image quality judgment device 6, the new surface image input unit 7 is input with new two-dimensional data obtained after a new reading of the planar object 1, and the new two-dimensional data is obtained by a one-dimensional line sensor. Figure 6 The image reading device 10 shown in (b) sequentially acquires new one-dimensional data in the sub-scanning direction. Then, the image quality determination unit 8 determines whether the planar object 1 corresponding to the new two-dimensional data is good or defective for each new one-dimensional data.

[0050] Here, the image quality judgment unit 8 can judge whether the planar object 1 corresponding to the new two-dimensional data is a good product or a defective product, and further extracts a basis area 9 corresponding to the existence area 1R in the new two-dimensional data as the basis for the judgment for each new one-dimensional data. In addition, through the control of the output unit 13 (sensor unit 12), each time the one-dimensional line sensor obtains new one-dimensional data, the new one-dimensional data is input into the new surface image input unit 7, so that the image quality judgment unit 8 can interrupt the judgment process at the moment when new one-dimensional data that is judged to be a defective product exists. Therefore, it is suitable for situations where you want to know about defective products as early as possible. In the case where there is a possibility that other parts of the planar object 1 (other than the basis area 9 for this judgment) also have the existence area 1R (based area 9) that can be judged as a defective product, the image quality judgment device 6 can be used again to judge the quality of the part that is later than the basis area 9 for this judgment in the sub-scanning direction.

[0051] Assume that in this case, by inputting short strips of image data (one-dimensional data) corresponding to various parts of the planar object 1 into the surface image input unit 3 as two-dimensional data, and allowing the learning unit 5 to individually learn whether a product is good or bad based on the short strips of image data (one-dimensional data), the image quality determination device 6 can determine whether the planar object 1 is good or bad without the learning unit 5 learning the overall image data of the planar object 1. This also includes multiple short strips of image data. For example, each part of the planar object 1 is a part that exists as a three-dimensional shape, including a pattern formed on the surface, a concave-convex portion formed on the surface, or a component mounted on the surface, or a part that exists as a color, including a pattern (color pattern), transparency, or a printed wiring pattern on the surface.

[0052] Thus far, the main image quality determination device (image reading device) of Embodiment 1 has been described with the main focus on determining the presence of an area 1R (based on area 9) in a planar object 1 that can be determined as a defective product. However, the main image quality determination device (image reading device) of Embodiment 1 can also determine the presence of an area 1R (based on area 9) that can be determined as a non-defective product. In addition, in the main image quality determination device (image reading device) of Embodiment 1, the so-called quality determination includes not only determining whether a product is good or defective, but also determining which area is a good product and which area is a defective product. In other words, when only one area of the planar object 1 is determined to be defective, the entire planar object 1 can be determined to be defective, or each area of the planar object 1 can be determined to be a good product or a defective product.

[0053] This is also the same as the learning of the learning unit 5 in the main image quality learning device of embodiment 1. In other words, the image quality learning device 2 can make the learning unit 5 learn in advance according to what quality judgment is performed by the image quality judgment device 6. In addition, if the scale of the learning unit 5 (learning model) can be increased, all changes in quality judgment can also be learned. That is, although Figure 3 Although one existing region 1R (based on region 9 ) is illustrated in FIG, a plurality of existing regions 1R (based on region 9 ) may exist on the image data.

[0054] In this case, linear one-dimensional data (short strip image data, a portion (one column) of the image data of the planar object 1) is input as two-dimensional data to the surface image input unit 3. This also includes a plurality of short strips of image data. In other words, the linear one-dimensional data is input as two-dimensional data to the surface image input unit 3. For example, consider inputting short strip image data (a portion (one column) of the image data of the planar object 1) obtained by at least one scanning in the sub-scanning direction intersecting the main scanning direction by a one-dimensional line sensor (image reading device 10) that reads the reading object (planar object 1) in the main scanning direction with the expansion direction of the linear one-dimensional data as the main scanning direction, into the surface image input unit 3. A portion (one column) of the image data of the planar object 1 refers to Figure 3 (B) and 3(C) show at least one column of image data.

[0055] As described above, the image quality learning device involved in embodiment 1, the image quality judgment device and the image reading device using the device can obtain two-dimensional data (including short strip image data, i.e. one-dimensional data) based on the planar object 1 and judgment information of good or defective products, and can provide an image quality learning device, an image quality judgment device and an image reading device that can provide a learning result (learning model) of the existence area of the three-dimensional shape or color of the surface of the planar object 1 in the two-dimensional data (including short strip image data, i.e. one-dimensional data) as the basis for judgment information of good or defective products, thereby learning the existence area as the basis for judgment information.

[0056] Description of labels

[0057] 1 Planar object, 1R existence area, 2 Image quality learning device, 3 Surface image input unit, 4 Judgment information input unit, 5 Learning unit, 6 Image quality judgment device, 7 New surface image input unit, 8 Image quality judgment unit, 9 Based on area, 10 Image reading device, 11 Optical unit, 11a Camera, 11b Lens array, 12 Sensor unit (sensor element array), 13 Output unit, 14 Light source (Line light source).

Claims

1. An image quality learning device for learning a judgment result of whether a planar object having at least one of a three-dimensional shape or a color on its surface is a good or bad product from image data of the surface of the planar object, characterized in that: include: a surface image input unit to which two-dimensional data as the image data of the surface of the planar object is input; a determination information input unit to which is input the determination information indicating whether the planar object corresponding to the two-dimensional data is a good product or a defective product; as well as a learning unit that learns, based on the input two-dimensional data and the input judgment information, an area where at least one of the three-dimensional shape or the color of the surface in the two-dimensional data exists, which serves as a basis for the judgment indicating whether the planar object corresponding to the two-dimensional data is a good product or a defective product, and the area where at least one of the three-dimensional shape or the color of the surface in the two-dimensional data exists is a two-dimensional area on the surface of the planar object. The surface image input unit inputs the two-dimensional data as the image data formed by arranging a plurality of linear one-dimensional data. The learning unit learns the position of the existing area separately for each of the one-dimensional data. By inputting short strip image data corresponding to each part of the planar object into the surface image input unit as two-dimensional data, and allowing the learning unit to learn good or bad products separately based on the short strip image data, The two-dimensional data formed by the one-dimensional data sequentially acquired in a sub-scanning direction intersecting the main scanning direction by a one-dimensional line sensor that reads a reading object in a main scanning direction is input to the surface image input unit.

2. The image quality learning device according to claim 1, wherein: The surface image input unit inputs the two-dimensional data, which is the image data of the surface of the planar object and has at least one of the pattern formed on the surface, the concave and convex parts formed on the surface, and the components mounted on the surface as the three-dimensional shape, or at least one of the pattern, transparency, and printed wiring pattern of the surface as the color.

3. The image quality learning device according to claim 1 or 2, characterized in that: The two-dimensional data includes information on the existing area and is input to the surface image input unit.

4. The image quality learning device according to claim 1 or 2, characterized in that: The determination information includes information on the existing area and is input into the determination information input unit.

5. The image quality learning device according to claim 1 or 2, characterized in that: The learning unit generates the determination information corresponding to the one-dimensional data from the determination information corresponding to the two-dimensional data, using the existing region of each one-dimensional data as a unit.

6. An image quality determination device using the learning results of the image quality learning device according to any one of claims 1 to 5, characterized in that: include: a new surface image input unit that inputs new two-dimensional data obtained by newly reading the planar object; as well as An image quality determination unit determines whether the planar object corresponding to the new two-dimensional data is a good product or a defective product based on a learning result learned by the learning unit.

7. The image quality determination device according to claim 6, wherein: The image quality determination unit determines whether the planar object corresponding to the new two-dimensional data is a good product or a defective product, and extracts a reference area corresponding to the existing area in the new two-dimensional data, which serves as a basis for the determination.

8. An image quality determination device, using the learning results of an image quality learning device, wherein the image quality learning device learns the determination result of whether a planar object having at least one of a three-dimensional shape or a color on its surface is a good or bad product from image data of the surface of the planar object, characterized in that: The image quality learning device includes: a surface image input unit to which two-dimensional data as the image data of the surface of the planar object is input; a determination information input unit to which is input the determination information indicating whether the planar object corresponding to the two-dimensional data is a good product or a defective product; and a learning unit that learns, based on the input two-dimensional data and the input judgment information, an area where at least one of the three-dimensional shape or the color of the surface in the two-dimensional data exists, which serves as a basis for the judgment indicating whether the planar object corresponding to the two-dimensional data is a good product or a defective product, and the area where at least one of the three-dimensional shape or the color of the surface in the two-dimensional data exists is a two-dimensional area on the surface of the planar object. The surface image input unit inputs the two-dimensional data obtained by arranging a plurality of one-dimensional data sequentially acquired in a sub-scanning direction intersecting the main scanning direction by a one-dimensional line sensor that reads a reading object in a main scanning direction. The image quality determination device comprises: a new surface image input unit to which new one-dimensional data obtained by newly reading the planar object is input; and an image quality determination unit that determines whether the planar object corresponding to the new one-dimensional data is a good product or a defective product based on the learning result learned by the learning unit; The new surface image input unit inputs the new one-dimensional data acquired by the one-dimensional line sensor in the sub-scanning direction. The image quality determination unit determines, for each of the new one-dimensional data, whether the planar object corresponding to the new one-dimensional data is a good product or a defective product. By inputting short strip image data corresponding to various parts of the planar object into the surface image input unit as two-dimensional data, the learning unit can learn good products or defective products individually based on the short strip image data.

9. The image quality determination device according to claim 8, wherein: The surface image input unit inputs two-dimensional data, which is the image data of the surface of the planar object and has at least one of a pattern formed on the surface, a concave-convex portion formed on the surface, and a component mounted on the surface as the three-dimensional shape, or at least one of a pattern, transparency, and a printed wiring pattern on the surface as the color.

10. The image quality judgment device according to claim 8 or 9, characterized in that: The two-dimensional data includes information on the existing area and is input to the surface image input unit.

11. The image quality judgment device according to claim 8 or 9, characterized in that: The determination information includes information on the existing area and is input into the determination information input unit.

12. An image quality judgment device using the learning results of the image quality learning device according to claim 1, characterized in that: include: a new surface image input unit that inputs new two-dimensional data obtained by newly reading the planar object; as well as an image quality determination unit configured to determine whether the planar object corresponding to the new two-dimensional data is a good product or a defective product based on the learning result learned by the learning unit; The new surface image input unit inputs the new two-dimensional data obtained by newly reading the planar object and consisting of new one-dimensional data sequentially acquired by the one-dimensional line sensor in the sub-scanning direction. The image quality determination unit determines, for each of the new one-dimensional data, whether the planar object corresponding to the new two-dimensional data is a good product or a defective product.

13. The image quality determination device according to claim 12, wherein: The image quality determination unit determines whether the planar object corresponding to the new two-dimensional data is a good product or a defective product, and extracts a reference area corresponding to the existing area in the new two-dimensional data as a basis for the determination for each new one-dimensional data.

14. The image quality determination device according to claim 12 or 13, wherein: Each time the one-dimensional line sensor acquires the new one-dimensional data, the newly acquired new one-dimensional data is input to the new surface image input unit. The image quality determination unit interrupts the determination process when the new one-dimensional data determined to be defective exists.

15. An image reading device comprising the image quality determination device according to any one of claims 6, 7, 12 to 14, characterized in that: include: an optical portion for converging light from the planar object, and A sensor section is configured to receive the light focused by the optical section and generate the new two-dimensional data.

16. The image reading device according to claim 15, wherein The optical part is a lens array in which a plurality of upright equal-magnification optical system lenses are arranged. The sensor unit is a sensor element array in which a plurality of sensor elements corresponding to each of the upright equal-magnification optical system lenses are arranged.

17. An image reading device comprising the image quality determination device according to any one of claims 8 to 11, characterized in that: include: an optical portion for converging light from the planar object, and A sensor portion is configured to receive the light focused by the optical portion and generate the new one-dimensional data.

18. The image reading device according to claim 17, wherein The optical part is a lens array in which a plurality of upright equal-magnification optical system lenses are arranged. The sensor unit is a sensor element array in which a plurality of sensor elements corresponding to each of the upright equal-magnification optical system lenses are arranged.

19. The image reading device according to claim 18, wherein The sensor element array outputs each of the new one-dimensional data to the new surface image input unit.

20. An image reading device comprising the one-dimensional line sensor of the image quality determination device according to any one of claims 12 to 14, characterized in that: include: a lens array having a plurality of upright equal-magnification optical system lenses arranged in the main scanning direction and converging light from the planar object; as well as A sensor element array is provided, wherein a plurality of sensor elements corresponding to each of the upright equal-magnification optical system lenses are arranged in the main scanning direction, and receives the light focused by the lens array to generate the new two-dimensional data.

21. The image reading device according to claim 20, wherein The sensor element array outputs each of the new one-dimensional data to the new surface image input unit.

Images