Foreign matter inspection device and foreign matter inspection method

Abstract

To provide a foreign matter inspection device and inspection method which can reduce cost.SOLUTION: A foreign matter inspection device 1 includes: an image acquisition part 231 for acquiring a first image IMG_UV photographed in a state in which an object TG is irradiated with ultraviolet light, a second image IMG_VL photographed in a state in which the object TG is irradiated with visible light, and a third image IMG_IR photographed in a state in which the object TG is irradiated with infrared light; an extraction part 232 for extracting a first extraction image IMG_EU of one channel from the first image, extracting a second extraction image IMG_EV of one channel from the second image, and extracting a third extraction image IMG_EI of one channel from the third image; a synthesis part 233 for synthesizing the first extraction image, the second extraction image, and the third extraction image to create a synthesis image IMG_CM; and a determination part 234 for determining a foreign matter using the synthesis image IMG_CM.SELECTED DRAWING: Figure 6

Description

[Technical field]

[0001] An embodiment of the present invention relates to a foreign matter inspection device and a foreign matter inspection method. [Background technology]

[0002] A deep learning model used in image inspection is known to be implemented and trained using the three channels of red, green, and blue from an RGB image captured using a visible light source as input.

[0003] It is known that a deep learning model trained using a large amount of image data for a certain task can achieve a relatively high score in other tasks with a relatively small amount of training data. This type of learning method is also called transfer learning.

[0004] For example, one of the foreign body inspections is an inspection to detect cartilage as a foreign body from meat. In the case of surface image inspection of meat, the accuracy of foreign body judgment may not be sufficient when inspecting images captured by a visible light source. For example, it is relatively easy to distinguish between red meat, fat, and cartilage. However, fat and cartilage are difficult to distinguish because of similar color tones. In such a case, image data captured using not only a visible light source but also light sources of other wavelengths can be used as input data for foreign body inspection. However, when multiple light sources are used to capture image data, the number of channels of image data used as input data increases. When the number of channels of image data is more than three, it is necessary to newly build a deep learning model corresponding to a special input. In that case, it is not possible to use the deep learning model corresponding to a three-channel input that was learned using the above-mentioned large amount of image data, and there is a risk that the cost of the foreign body inspection device will increase by building a new model. [Prior art documents] [Patent documents]

[0005] [Patent Document 1] JP 2018-066649 A Summary of the Invention [Problem to be solved by the invention]

[0006] An object of the present invention is to provide a foreign matter inspection apparatus and inspection method capable of reducing costs. [Means for solving the problem]

[0007] According to an embodiment, the foreign body inspection device includes an image acquisition unit that acquires a first image captured while an object is irradiated with visible light, a second image captured while an object is irradiated with ultraviolet light, and a third image captured while an object is irradiated with infrared light, an extraction unit that extracts a first extracted image of one channel from the first image, extracts a second extracted image of one channel from the second image, and extracts a third extracted image of one channel from the third image, a synthesis unit that synthesizes the first extracted image, the second extracted image, and the third extracted image to generate a synthetic image, and a determination unit that determines a foreign body in the object using the synthetic image. Effect of the Invention

[0008] According to the present invention, it is possible to provide a foreign matter inspection device and a foreign matter inspection method that can reduce costs. [Brief description of the drawings]

[0009] [Figure 1] FIG. 1 is a block diagram showing an example of the configuration of a foreign matter inspection device according to the first embodiment. [Diagram 2] FIG. 2 is a block diagram showing an example of a hardware configuration of the foreign matter inspection apparatus according to the first embodiment. [Diagram 3] FIG. 3 is a schematic diagram showing an example of an ultraviolet light source included in the light source unit of the foreign matter inspection device according to the first embodiment. [Figure 4] FIG. 4 is a block diagram showing an example of the functional configuration of a processing unit included in the foreign matter inspection apparatus according to the first embodiment. [Diagram 5]FIG. 5 is a block diagram showing an example of the functional configuration of the determination processing unit and the configuration of the storage unit included in the foreign matter inspection apparatus according to the first embodiment. [Figure 6] FIG. 6 is a conceptual diagram showing a flow of synthetic image generation in the foreign matter inspection device according to the first embodiment. [Figure 7] FIG. 7 is a diagram showing a first specific example of a captured image, an extracted image, and a composite image in the foreign matter inspection device according to the first embodiment. [Figure 8] FIG. 8 is a diagram showing a second specific example of the captured image, the extracted image, and the composite image in the foreign matter inspection device according to the first embodiment. [Figure 9] FIG. 9 is a flowchart showing an example of the flow of the foreign object determination process according to the first embodiment. [Figure 10] FIG. 10 is a block diagram showing an example of the configuration of a foreign matter inspection device according to the second embodiment. [Figure 11] FIG. 11 is a block diagram showing an example of a hardware configuration of a foreign matter inspection apparatus according to the second embodiment. [Figure 12] FIG. 12 is a schematic diagram showing an example of an ultraviolet light source included in the light source unit of the foreign matter inspection device according to the second embodiment. [Figure 13] FIG. 13 is a block diagram showing an example of the hardware configuration of a foreign matter inspection apparatus according to the third embodiment. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0010] The foreign body inspection device (1) of the embodiment includes an image acquisition unit (231) that acquires a first image (IMG_UV) captured while irradiating an object (TG) with ultraviolet light, a second image (IMG_VL) captured while irradiating the object (TG) with visible light, and a third image (IMG_IR) captured while irradiating the object (TG) with infrared light, and extracts a first extracted image (IMG_EU) of one channel from the first image (IMG_UV) and a third extracted image (IMG_IR) of one channel from the second image (IMG_VL). The inspection device (1) includes an extraction unit (232) that extracts a second extracted image (IMG_EV) of one channel from the third image (IMG_IR) and extracts a third extracted image (IMG_EI) of one channel from the third image (IMG_IR), a synthesis unit (233) that synthesizes the first extracted image (IMG_EU), the second extracted image (IMG_EV), and the third extracted image (IMG_EI) to generate a synthesis image (IMG_CM), and a determination unit (234) that uses the synthesis image (IMG_CM) to determine whether a foreign object is present in the target (TG). This allows the foreign object inspection device (1) to perform foreign object inspection using the synthesis image (IMG_CM) of the three channels as an input.

[0011] The determination unit (234) of the foreign body inspection device (1) of the embodiment is configured to perform foreign body determination by using a deep learning model (303) with the composite image (IMG_CM) as input. This allows the foreign body inspection device (1) to use the deep learning model (303) with the three-channel composite image (IMG_CM) as input. Therefore, the foreign body inspection device (1) can reduce the cost of building a new deep learning model.

[0012] The first extracted image (IMG_EU) is an image of the B channel of the first image (IMG_UV), the second extracted image (IMG_EV) is an image of the R channel of the second image (IMG_VL), and the third extracted image (IMG_EI) is a grayscale image of the third image (IMG_IR). This allows the foreign body inspection device (1) to extract an extracted image of one channel that is effective for identifying foreign bodies from each captured image.

[0013] The foreign body inspection device (1) of the embodiment further includes a light source unit (12) configured to be able to irradiate an object (TG) with ultraviolet light, visible light, and infrared light, and an imaging unit (13) that images the object (TG). This allows the foreign body inspection device (1) to capture a first image (IMG_UV), a second image (IMG_VL), and a third image (IMG_IR) of the object (TG).

[0014] The foreign body inspection method of the embodiment includes acquiring (S2) a first image (IMG_UV) captured while irradiating an object (TG) with ultraviolet light, a second image (IMG_VL) captured while irradiating the object (TG) with visible light, and a third image (IMG_IR) captured while irradiating the object (TG) with infrared light, extracting (S3) a first extracted image (IMG_EU) of one channel from the first image (IMG_UV), and extracting (S4) a second extracted image (IMG_EU) of one channel from the second image (IMG_VL). The method includes extracting a second extracted image (IMG_EV) of the first channel from the third image (IMG_IR) (S4), extracting a third extracted image (IMG_EI) of one channel from the third image (IMG_IR) (S5), synthesizing the first extracted image (IMG_EU), the second extracted image (IMG_EV), and the third extracted image (IMG_EI) to generate a composite image (IMG_CM) of three channels (S6), and determining whether or not there is a foreign object in the target object (TG) using the composite image (IMG_CM) (S7). This allows the foreign object inspection device (1) to perform foreign object inspection using the composite image (IMG_CM) of the three channels as an input.

[0015] The foreign body inspection method of the embodiment uses a deep learning model (303) with a composite image (IMG_CM) as input. This allows the foreign body inspection device (1) to use the deep learning model (303) with a three-channel composite image (IMG_CM) as input. Therefore, the foreign body inspection device (1) can reduce the cost of building a new deep learning model.

[0016] Hereinafter, an embodiment will be described with reference to the drawings. In the following embodiment, a case will be described in which the object of foreign body inspection is meat, and cartilage contained in the meat is detected as a foreign body. Note that the object of inspection is not limited to meat. The object of inspection may be other food products or non-food products.

[0017] 1. First embodiment The foreign matter inspection device according to the first embodiment will be described.

[0018] 1.1 Configuration of foreign body inspection equipment The configuration of a foreign matter inspection device according to this embodiment will be described with reference to Fig. 1. Fig. 1 is a block diagram showing an example of the configuration of a foreign matter inspection device.

[0019] The foreign body inspection device 1 is a device that inspects an object for the presence or absence of foreign bodies. Hereinafter, the object inspected by the foreign body inspection device 1 will be referred to as "object TG". The object TG is, for example, dressed meat. Dressed meat is, for example, beef, pork, chicken, and the like. Dressed meat is broadly divided into three types of parts: lean meat, cartilage, and fat. Dressed meat has, for example, a block shape. Hereinafter, the dressed meat inspected by the foreign body inspection device 1 will be referred to as "dressed meat TG1". For example, the foreign body inspection device 1 inspects the dressed meat TG1 for the presence or absence of foreign bodies (cartilage).

[0020] As shown in FIG. 1, the foreign matter inspection device 1 includes a processing unit 11, a light source unit 12, and an imaging unit 13.

[0021] The processing unit 11 is, for example, a computer. The processing unit 11 transmits control commands to the light source unit 12 and the photographing unit 13, and controls the operations of the light source unit 12 and the photographing unit 13. The processing unit 11 also acquires an image captured by the photographing unit 13. The processing unit 11 determines the presence or absence of a foreign object in the target TG based on the acquired image. Hereinafter, the image captured by the photographing unit 13 is also referred to as a "captured image."

[0022] The light source unit 12 is, for example, an LED (Light Emitting Diode). The light source unit 12 irradiates ultraviolet light, visible light, and infrared light based on a control command received from the processing unit 11. In other words, the light source unit 12 can supply ultraviolet light, visible light, and infrared light. For example, the light source unit 12 irradiates the target object TG with ultraviolet light, visible light, and infrared light in that order. The light source unit 12 may be configured to irradiate each of the ultraviolet light, visible light, and infrared light alone, or any two of them in combination.

[0023] The photographing unit 13 is, for example, a camera, and is capable of photographing images and videos. In the following, the photographing unit 13 will be described as photographing an image. The photographing unit 13 photographs an image based on a control command received from the processing unit 11. For example, the photographing unit 13 photographs an image of the object TG in each (or in any one) of the following states: ultraviolet light is irradiated onto the object TG by the light source unit 12 and the ultraviolet light is reflected by the object TG; visible light is irradiated onto the object TG by the light source unit 12 and the visible light is reflected by the object TG; and infrared light is irradiated onto the object TG by the light source unit 12 and the infrared light is reflected by the object TG.

[0024] 1.1.1 Hardware configuration of the foreign body inspection device The hardware configuration of the foreign matter inspection device 1 will be described with reference to FIG. 2. FIG. 2 is a block diagram showing an example of the hardware configuration of the foreign matter inspection device 1. FIG. 2 shows a situation in which fresh meat TG1 placed on a conveying section CV and conveyed in the direction of the arrow (from upstream to downstream in the conveying direction of the conveying section CV) is inspected by the foreign matter inspection device 1. The conveying section CV is, for example, a conveyor. For example, in the conveying section CV, multiple pieces of fresh meat TG1 of the same type are conveyed with a predetermined interval between them in the conveying direction of the conveying section CV. The foreign matter inspection device 1 then sequentially inspects the multiple pieces of fresh meat TG1 one by one.

[0025] As shown in FIG. 2, the processing unit 11 includes a processor 111, a storage medium 112, a communication interface 113, and a user interface 114.

[0026] The processor 111 is, for example, a central processing unit (CPU). The processor 111 may be any one of a graphics processing unit (GPU), an application specific integrated circuit (ASIC), a microcomputer, a field programmable gate array (FPGA), and a digital signal processor (DSP), or may be a combination of these. The processor 111 executes, for example, a process of controlling the operations of the light source unit 12 and the photographing unit 13, and a process of determining whether or not a foreign object is present in the object TG. Hereinafter, the process of controlling the operations of the light source unit 12 and the photographing unit 13 is referred to as an "operation control process". The process of determining whether or not a foreign object is present in the object TG is referred to as a "foreign object determination process". The details of the operation control process and the foreign object determination process will be described later.

[0027] The storage medium 112 includes, for example, a read only memory (ROM) and a random access memory (RAM). The ROM stores, for example, a program for causing the processor 111 to execute an operation control process and a foreign object determination process. The RAM is used as a working area for the processor 111. The RAM temporarily stores, for example, the above-mentioned program executed by the processor 111 and data at the time of execution of the foreign object determination process. The storage medium 112 also stores a deep learning model used in foreign object inspection and learning data used for learning the deep learning model. Note that the above-mentioned program and deep learning model executed by the processor 111 may be stored in a computer (server) connected via a network such as the Internet, or a server in a cloud environment. In this case, the program and deep learning model are downloaded by the processor 111 via the network.

[0028] The communication interface 113 is responsible for communication between the light source unit 12, the photographing unit 13, and other external devices or apparatuses and the processing unit 11. The communication interface 113 transmits, for example, control commands generated by the processing unit 11 and control commands received from the outside to the light source unit 12 and the photographing unit 13. The communication interface 113 receives, for example, captured images from the photographing unit 13. The communication interface 113 can receive learning data of deep learning from the outside. In addition, the communication interface 113 can transmit, for example, a determination result (determination information) obtained by the foreign object determination process to other external devices or apparatuses.

[0029] The user interface 114 is a means for a user of the foreign matter inspection device 1 to input operations related to the inspection. The user interface 114 is, for example, a button, a switch, or a touch panel. The user interface 114 may be provided separately from the processing unit 11.

[0030] The light source unit 12 includes an ultraviolet light source 121, a visible light source 122, and an infrared light source 123. The ultraviolet light source 121, the visible light source 122, and the infrared light source 123 are arranged side by side in the conveying direction of the conveying unit CV. The ultraviolet light source 121 includes one or more ultraviolet LEDs. The ultraviolet light irradiated by the ultraviolet light source 121 is ultraviolet light having a peak wavelength, for example, any wavelength in the range of 365 nm ± 30 nm. The visible light source 122 includes one or more visible light LEDs. The infrared light source 123 includes one or more infrared light LEDs. The infrared light irradiated by the infrared light source 123 is infrared light having a peak wavelength, for example, any wavelength in the range of 1300 nm ± 30 nm. The ultraviolet light LED in the ultraviolet light source 121, the visible light LED in the visible light source 122, and the infrared light LED in the infrared light source 123 are fixed to the lower surface of the support unit SP. The irradiation direction of the ultraviolet light LED, the visible light LED, and the infrared light LED is the opposite direction to the support part SP, i.e., the direction facing the transport part CV. Note that the light sources provided in the ultraviolet light source 121, the visible light source 122, and the infrared light source 123 do not have to be LEDs, and may be, for example, non-solid light emitting materials such as lamps, or other solid light emitting materials (such as laser diodes and organic EL).

[0031] 2, the ultraviolet light source 121, the visible light source 122, and the infrared light source 123 are arranged in the order of the ultraviolet light source 121, the visible light source 122, and the infrared light source 123 from the upstream side in the transport direction of the transport unit CV, but the arrangement order of the ultraviolet light source 121, the visible light source 122, and the infrared light source 123 is not limited thereto. For example, the ultraviolet light source 121, the visible light source 122, and the infrared light source 123 may be arranged in the order of the visible light source 122, the ultraviolet light source 121, and the infrared light source 123 from the upstream side in the transport direction of the transport unit CV.

[0032] The photographing unit 13 includes an ultraviolet camera 131, a visible light camera 132, and an infrared camera 133. The ultraviolet camera 131 photographs an image of the object TG in a state where the object TG is irradiated with ultraviolet light. The visible light camera 132 photographs an image of the object TG in a state where the object TG is irradiated with visible light. The infrared camera 133 photographs an image of the object TG in a state where the object TG is irradiated with infrared light. Hereinafter, the ultraviolet camera 131 is also referred to as a "first photographing unit". The visible light camera 132 is also referred to as a "second photographing unit". The infrared camera 133 is also referred to as a "third photographing unit". For example, the ultraviolet camera 131, the visible light camera 132, and the infrared camera 133 are arranged side by side in the conveying direction of the conveying unit CV. The ultraviolet camera 131, the visible light camera 132, and the infrared camera 133 are fixed to the lower surface of the support unit SP. The imaging directions of the ultraviolet camera 131, the visible light camera 132, and the infrared camera 133 are opposite to the supporting part SP, that is, facing the conveying part CV.

[0033] FIG. 3 is a schematic diagram showing an example of the ultraviolet light source 121. FIG. 3 shows the ultraviolet light source 121 as viewed from the transport section CV side (i.e., the target TG side). As shown in FIG. 3, the ultraviolet light source 121 has a plurality of ultraviolet LEDs arranged in a ring shape as viewed from the transport section CV side. The ultraviolet camera 131 is arranged so as to be surrounded by the plurality of ultraviolet LEDs as viewed from the transport section CV side. The visible light LEDs in the visible light source 122 and the infrared light LEDs in the infrared light source 123 may also be arranged in a ring shape as viewed from the transport section CV side, similar to the ultraviolet light source 121. In this case, the visible light camera 132 is arranged so as to be surrounded by the plurality of visible light LEDs as viewed from the transport section CV side. The infrared camera 133 is arranged so as to be surrounded by the plurality of infrared LEDs as viewed from the transport section CV side.

[0034] 3, the multiple ultraviolet LEDs in the ultraviolet light source 121 are arranged in a ring shape when viewed from the transport unit CV side, but the arrangement of the multiple ultraviolet LEDs in the ultraviolet light source 121 is not limited to this. For example, the multiple ultraviolet LEDs in the ultraviolet light source 121 may be arranged side by side in the transport direction on the transport unit CV side when viewed from the transport unit CV side. The same applies to the multiple visible light LEDs in the visible light source 122 and the multiple infrared light LEDs in the infrared light source 123.

[0035] 1.1.2 Functional configuration of the processing section Next, the functional configuration of the processing unit 11 will be described with reference to Fig. 4. Fig. 4 is a block diagram showing an example of the functional configuration of the processing unit 11. Fig. 4 also shows the light source unit 12 and the image capturing unit 13.

[0036] The processing unit 11 includes, as functional blocks, a process execution unit 211 and a storage unit 212. The processor 111 of the processing unit 11 functions as the process execution unit 211. The storage medium 112 of the processing unit 11 functions as the storage unit 212.

[0037] The process execution unit 211 executes the operation control process and the foreign object determination process. The process execution unit 211 includes, as functional blocks, an operation control unit 221 and a determination processing unit 222. The processor 111 of the processing unit 11 executes a program to function as the operation control unit 221 and the determination processing unit 222. The operation control unit 221 and the determination processing unit 222 each perform a part of the process performed by the process execution unit 211.

[0038] (Operation control unit 221) The operation control unit 221 transmits control commands to each of the ultraviolet light source 121, the visible light source 122, and the infrared light source 123, and controls the operations of each of the ultraviolet light source 121, the visible light source 122, and the infrared light source 123. That is, the operations of each of the ultraviolet light source 121, the visible light source 122, and the infrared light source 123 are controlled by the operation control unit 221. The operation control unit 221 also transmits control commands to each of the ultraviolet camera 131, the visible light camera 132, and the infrared camera 133, and controls the operations of each of the ultraviolet camera 131, the visible light camera 132, and the infrared camera 133. That is, the operations of each of the ultraviolet camera 131, the visible light camera 132, and the infrared camera 133 are controlled by the operation control unit 221.

[0039] Under the control of the operation control unit 221, the photographing unit 13 photographs the dressed meat TG1 in each of the states where the ultraviolet light source 121 irradiates the dressed meat TG1 with ultraviolet light, the visible light source 122 irradiates the dressed meat TG1 with visible light, and the infrared light source 123 irradiates the dressed meat TG1 with infrared light. That is, the ultraviolet camera 131 photographs the dressed meat TG1 in a state where the dressed meat TG1 is irradiated with ultraviolet light. The visible light camera 132 photographs the dressed meat TG1 in a state where the dressed meat TG1 is irradiated with visible light. The infrared camera 133 photographs the dressed meat TG1 in a state where the dressed meat TG1 is irradiated with infrared light. As a result, photographed images of the dressed meat TG1 are generated in each of the states where only ultraviolet light is irradiated, only visible light is irradiated, and only infrared light is irradiated. Hereinafter, the image of the dressed meat TG1 photographed by the ultraviolet camera 131 in a state where ultraviolet light is irradiated is referred to as an "ultraviolet image." An image of the fresh meat TG1 captured by the visible light camera 132 under visible light irradiation is referred to as a "visible light image." An image of the fresh meat TG1 captured by the infrared camera 133 under infrared light irradiation is referred to as an "infrared light image."

[0040] In one example, in a state where the fresh meat TG1 is moving from the upstream side to the downstream side in the conveying direction of the conveyor CV, the operation control unit 221 causes the light source unit 12 to sequentially irradiate the fresh meat TG1 with ultraviolet light, visible light, and infrared light. Then, in each of the states where the ultraviolet light is irradiated, the visible light is irradiated, and the infrared light is irradiated, the operation control unit 221 causes the photographing unit 13 to photograph the fresh meat TG1. Note that the order in which the light source unit 12 irradiates the fresh meat TG1 with ultraviolet light, visible light, and infrared light does not have to be ultraviolet light, visible light, and infrared light. Infrared light has a property of generating heat. Therefore, when the fresh meat TG1 is irradiated with infrared light, the moisture state of the surface of the fresh meat TG1 may change. This may affect the photographing in the state where the ultraviolet light is irradiated and the state where the visible light is irradiated. For this reason, it is preferable that the light source unit 12 irradiates the fresh meat TG1 with infrared light last. That is, it is preferable that the light source unit 12 irradiates the fresh meat TG1 with ultraviolet light and visible light, and then irradiates the fresh meat TG1 with infrared light.

[0041] (Properties of each part of meat in relation to visible, ultraviolet, and infrared light) The characteristics of red meat, cartilage, and fat when meat TG1 is irradiated with visible light, ultraviolet light, and infrared light will be described.

[0042] In an environment where the fresh meat TG1 is irradiated with visible light, the red meat, cartilage, and fat have different appearances, such as color. For this reason, for example, in a visible light image obtained by irradiating visible light, by detecting parts that are different in color, etc. from the red meat, it becomes possible to distinguish the red meat from the cartilage and fat. For example, in the visible light image, the parts other than red that are included in the area where the fresh meat TG1 is located can be distinguished as cartilage and fat.

[0043] In addition, in an environment where visible light is irradiated onto the fresh meat TG1, the cartilage and fat have similar appearances, such as color, making it difficult to distinguish between cartilage and fat from only the visible light image.

[0044] In an environment where the fresh meat TG1 is irradiated with ultraviolet light in a predetermined wavelength range, the cartilage and fat tend to fluoresce (emit) in a predetermined color (for example, white or yellow) like a fluorescent material. In particular, by irradiating the fresh meat TG1 containing a mixture of cartilage and fat with ultraviolet light having a peak wavelength of 365 nm or a wavelength in the vicinity thereof, the intensity of light emitted from the fluorescent cartilage and fat increases, and the brightness of the fluorescent cartilage and fat tends to increase. Even if the fresh meat TG1 is irradiated with ultraviolet light in a predetermined wavelength range in which the cartilage and fat tend to fluoresce, the red meat does not fluoresce and appears blackish. For this reason, for example, by detecting the fluorescent parts in an ultraviolet image obtained by irradiating ultraviolet light having a peak wavelength in any wavelength range of 365 nm ± 30 nm, it is possible to distinguish the red meat from the cartilage and fat. For example, it is possible to distinguish the parts with relatively high brightness in the range in which the fresh meat TG1 is located in the ultraviolet image as cartilage and fat. It is preferable that the ultraviolet light irradiated by the light source unit 12 has a peak wavelength of 365 nm.

[0045] In addition, in an environment where the fresh meat TG1 is irradiated with ultraviolet light in a specific wavelength range, not only the cartilage but also the fat is likely to fluoresce, making it difficult to distinguish between cartilage and fat from the ultraviolet image alone.

[0046] In an environment where the fresh meat TG1 is irradiated with infrared light in a predetermined wavelength range, the cartilage and red meat tend to absorb the infrared light. Therefore, the amount of infrared light reflected by the cartilage and red meat is relatively small, and the cartilage and red meat appear black. In particular, by irradiating the fresh meat TG1 containing the cartilage and red meat with infrared light having a peak wavelength of 1300 nm or a wavelength in the vicinity thereof, the intensity of light emitted from the cartilage and red meat that have absorbed the infrared light is reduced, and the brightness of the cartilage and red meat that have absorbed the infrared light tends to be reduced. Even if the fresh meat TG1 is irradiated with infrared light in a predetermined wavelength range that is easily absorbed by the cartilage and red meat, the fat does not absorb the infrared light. Therefore, for example, by detecting black parts in an infrared light image obtained by irradiating infrared light having a peak wavelength in any wavelength range of 1300 nm ± 30 nm, it is possible to distinguish fat from cartilage and red meat. For example, in the infrared light image, a portion with a relatively low brightness within the range in which the fresh meat TG1 is located can be identified as cartilage and red meat. Note that the infrared light irradiated by the light source unit 12 preferably has a peak wavelength at 1300 nm.

[0047] In addition, in an environment where the fresh meat TG1 is irradiated with infrared light in a specific wavelength range, not only the cartilage but also the red meat is likely to absorb the infrared light, making it difficult to distinguish between the cartilage and the red meat from the infrared light image alone.

[0048] (Determination processing unit 222) The determination processing unit 222 determines the presence or absence of a foreign object contained in the target object TG. That is, the determination processing unit 222 determines the presence or absence of cartilage in the dressed meat TG1. The determination processing unit 222 uses, for example, a deep learning model that is compatible with transfer learning and optimized for three-channel input, for determining cartilage. The determination processing unit 222 extracts an image of one channel from each of the visible light image, the ultraviolet light image, and the infrared light image as a preprocessing of data input to the deep learning model. Hereinafter, the extracted image is also referred to as an "extracted image". Then, the determination processing unit 222 generates a three-channel composite image by combining the three extracted images. Therefore, the images of the three channels of the composite image have different imaging conditions (light sources). The determination processing unit 222 uses the composite image as input data for the deep learning model. That is, the determination processing unit 222 executes a process of determining the presence or absence of cartilage using the composite image.

[0049] 1.1.3 Functional configuration of the judgment processing unit Next, an example of the functional configuration of the determination processing unit 222 will be described with reference to Fig. 5. Fig. 5 is a block diagram showing an example of the functional configuration of the determination processing unit 222 and the configuration of the storage unit 212.

[0050] 5, the determination processing unit 222 includes an image acquiring unit 231, an image extracting unit 232, an image combining unit 233, a foreign substance determination unit 234, a determination information generating unit 235, a learning data acquiring unit 236, and a learning processing unit 237. The functions of the image acquiring unit 231, the image extracting unit 232, the image combining unit 233, the foreign substance determination unit 234, the determination information generating unit 235, the learning data acquiring unit 236, and the learning processing unit 237 are realized by a program executed by the processor 111.

[0051] (Image acquisition unit 231) The image acquiring section 231 acquires image data captured by the photographing section 13. More specifically, the image acquiring section 231 acquires a visible light image, an ultraviolet light image, and an infrared light image of the dressed meat TG1 captured by the photographing section 13. The image acquiring section 231 may perform filtering such as noise removal or image quality adjustment on the acquired image data. The image acquiring section 231 transmits the acquired visible light image, ultraviolet light image, and infrared light image to the image extraction section 232. Note that the storage section 212 may store the visible light image, ultraviolet light image, and infrared light image acquired by the image acquiring section 231 as one of the inspection data 301.

[0052] (Image extraction unit 232) The image extraction unit 232 extracts an extracted image of one channel from each of the visible light image, the ultraviolet light image, and the infrared light image. That is, the image extraction unit 232 extracts three extracted images. The image extraction unit 232 may extract an extracted image of one channel that is most effective for identifying a foreign object from each captured image. More specifically, the visible light image is an RGB image. The image extraction unit 232 extracts, for example, an extracted image of the R (red) channel from the visible light image. The image extraction unit 232 may extract an extracted image of the B (blue) channel or the G (green) channel from the visible light image. The ultraviolet light image is an RGB image. The image extraction unit 232 extracts, for example, an extracted image of the B channel from the ultraviolet light image. The image extraction unit 232 may extract an extracted image of the R channel or the G channel from the ultraviolet light image. The infrared light image is a grayscale image of one channel. The image extraction unit 232 extracts a grayscale extracted image from the infrared image. That is, the image extraction unit 232 uses the infrared image as it is as the extracted image. The image extraction unit 232 transmits the three extracted images to the image synthesis unit 233. The storage unit 212 may store the three extracted images as one of the test data 301.

[0053] (Image synthesis unit 233) The image synthesis unit 233 synthesizes the three extracted images to generate a three-channel synthetic image. The image synthesis unit 233 transmits the synthetic image to the foreign matter determination unit 234. The storage unit 212 may store the synthetic image as one of the inspection data 301.

[0054] (Foreign object determination unit 234) The foreign object determination unit 234 reads out the deep learning model 303 from the storage unit 212. Then, the foreign object determination unit 234 determines whether or not the dressed meat TG1 contains cartilage by applying the composite image as an input to the deep learning model 303. The foreign object determination unit 234 transmits the determination result to the determination information generation unit 235.

[0055] (Judgment information generation unit 235) The determination information generating section 235 generates determination information based on the determination result in the foreign matter determining section 234. Then, the determination information generating section 235 outputs the generated determination information from the determination processing section 222. Note that the storage section 212 may store the determination information as one of the inspection data 301.

[0056] In one example, the determination information is notified to a user of the foreign matter inspection device 1, etc., via a display device (not shown). For example, the determination information indicates whether or not cartilage (foreign matter) is present in the fresh meat TG1. In this case, the display device displays flags such as "cartilage present" and "no cartilage." In addition, for example, the determination information indicates position information of the cartilage present in the fresh meat TG1. In this case, the position information of the cartilage may be displayed as numerical values ​​such as position coordinates, or may be displayed in image format. In the case of image format, the position information of the cartilage present in the fresh meat TG1 is displayed, for example, by coloring or surrounding the part corresponding to the cartilage in the visible light image.

[0057] In one example, the determination information generating unit 235 generates, as output information, an operation command (control command) for a removal device (not shown) that removes cartilage. The determination information generating unit 235 then outputs the operation command generated as the determination information to the removal device. The removal device can remove cartilage from the dressed meat TG1 based on the determination information.

[0058] (Learning data acquisition unit 236) The learning data acquisition unit 236 acquires learning data 302 used in the learning process performed by the learning processing unit 237. The learning data 302 may be a composite image that has already been determined as a foreign substance, or may be an undetermined composite image. The learning data 302 may be, for example, a composite image previously generated by the foreign substance inspection device 1, or may be input by a user or the like via the user interface 114, or may be input from an external device via the communication interface 113. The learning data acquisition unit 236 transmits the acquired learning data 302 to the learning processing unit 237. The learning data acquisition unit 236 also causes the storage unit 212 to store the learning data 302.

[0059] (Learning processing unit 237) The learning processing unit 237 executes learning processing using the learning data 302 and updates the deep learning model 303. The learning processing unit 237 stores the updated deep learning model 303 in the storage unit 212. The learning processing unit 237 executes deep learning by a neural network as the learning processing. The deep learning model 303 is a model optimized for three-channel input. For example, the deep learning model 303 that has been trained using a large amount of image data other than the synthetic image is used for the deep learning model 303. The learning processing unit 237 executes transfer learning using the learning data 302 for the deep learning model 303. For example, the deep learning model 303 that has been trained using a large amount of image data has undergone adjustment of parameters in the intermediate layer of the neural network. For example, by performing transfer learning using the learning data 302 in some layers of the intermediate layer, a relatively high score (test accuracy) can be obtained with a relatively small amount of data. The learning processing unit 237 may be configured to perform learning using the learning data 302 on the deep learning model 303 in a state where learning has not been performed using image data other than the synthetic image.

[0060] It should be noted that the determination processing unit 222 may not include the learning processing unit 237. In this case, the processing unit 11 receives the updated deep learning model 303 from an external deep learning model learning server or the like.

[0061] 1.2 Example of synthetic image generation Next, specific examples of composite image generation will be described with reference to Figs. 6 to 8. Fig. 6 is a conceptual diagram showing the flow of composite image generation. Fig. 7 is a diagram showing a first specific example of a photographed image, an extracted image, and a composite image. Fig. 8 is a diagram showing a second specific example of a photographed image, an extracted image, and a composite image. Figs. 7 and 8 show a case where the dressed meat TG1 is chicken. In Figs. 7 and 8, the outline of the dressed meat TG1 is shown by a white line to clarify the outline of the dressed meat TG1.

[0062] As shown in FIG. 6, the image acquisition unit 231 acquires an ultraviolet light image IMG_UV, a visible light image IMG_VL, and an infrared light image IMG_IR from the imaging unit 13.

[0063] 7 and 8, in the ultraviolet light image IMG_UV, the brightness of cartilage and fat is high, and the brightness of red meat is low. When attention is paid to the gradation value of the B channel of the ultraviolet light image IMG_UV, it is relatively easy to distinguish between red meat, cartilage, and fat. For this reason, the image extraction unit 232 extracts, for example, an extracted image IMG_EU of the B channel from the ultraviolet light image IMG_UV.

[0064] In the visible light image IMG_VL, there is a difference in vividness of red between the red meat of the dressed meat TG1 and the cartilage and fat. When attention is paid to the gradation value of the R channel of the visible light image IMG_VL, it is relatively easy to distinguish the red meat from the cartilage and fat. For this reason, the image extraction unit 232 extracts, for example, an R channel extraction image IMG_EV from the visible light image IMG_VL.

[0065] In the infrared image IMG_IR, the luminance values ​​of red meat and cartilage are lower than the luminance value of fat. Therefore, in the infrared image IMG_IR, it is relatively easy to distinguish fat from red meat and cartilage. The image extraction unit 232 extracts the infrared image IMG_IR as a grayscale extracted image IMG_EI.

[0066] As shown in FIGS. 6 to 8, the image synthesis unit 233 synthesizes the extraction images IMG_EU, IMG_EV, and IMG_EI to generate a three-channel synthetic image IMG_CM.

[0067] 1.3 Flow of foreign object detection process Next, an example of the flow of a foreign substance determination process will be described with reference to Fig. 9. Fig. 9 is a flowchart showing an example of the flow of a foreign substance determination process.

[0068] 9, when the determination processing unit 222 (foreign object determination unit 234) starts the foreign object determination process, it reads out the deep learning model 303 from the storage unit 212 (S1). The deep learning model 303 is the latest model learned by the learning processing unit 237.

[0069] The image acquisition unit 231 acquires the ultraviolet light image IMG_UV, the visible light image IMG_VL, and the infrared light image IMG_IR of the dressed meat TG1 from the photographing unit 13 (S2). The image acquisition unit 231 transmits the ultraviolet light image IMG_UV, the visible light image IMG_VL, and the infrared light image IMG_IR to the image extraction unit 232.

[0070] Next, the image extraction unit 232 extracts three extracted images. More specifically, the image extraction unit 232 extracts an extracted image IMG_EU of one channel (B channel) from the ultraviolet light image IMG_UV (S3). The image extraction unit 232 extracts an extracted image IMG_EV of one channel (R channel) from the visible light image IMG_VL (S4). The image extraction unit 232 also extracts an extracted image IMG_EI of one channel (grayscale) from the infrared light image IMG_IR (S5). The image extraction unit 232 may change the order of the processes of steps S3 to S5, or may execute them simultaneously. The image extraction unit 232 transmits the extracted images IMG_EU, IMG_EV, and IMG_EI to the image synthesis unit 233.

[0071] Next, the image synthesis unit 233 synthesizes the three extracted images IMG_EU, IMG_EV, and IMG_EI to generate a three-channel synthetic image IMG_CM (S6). The image synthesis unit 233 transmits the synthetic image IMG_CM to the foreign object determination unit 234.

[0072] Next, the foreign object determination unit 234 executes a foreign object determination using the deep learning model 303 with the composite image IMG_CM as input (S7). That is, the foreign object determination unit 234 determines whether or not cartilage is contained in the dressed meat TG1 in the composite image IMG_CM. The foreign object determination unit 234 transmits the determination result to the determination information generation unit 235.

[0073] Next, the determination information generating unit 235 outputs the determination result (S8). More specifically, the determination information generating unit 235 generates the determination information based on the determination result. Then, the determination information generating unit 235 outputs the determination information to the outside.

[0074] 1.4 Effects of this embodiment In the configuration according to this embodiment, the foreign body inspection device 1 can capture an ultraviolet light image IMG_UV, a visible light image IMG_VL, and an infrared light image IMG_IR of the fresh meat TG1 (target object TG). The foreign body inspection device 1 can extract an extracted image of one channel that is effective for identifying foreign bodies from each of the ultraviolet light image IMG_UV, the visible light image IMG_VL, and the infrared light image IMG_IR. The foreign body inspection device 1 can generate a three-channel composite image IMG_CM by combining the three extracted images. The foreign body inspection device 1 can use the three-channel composite image IMG_CM as an input for the deep learning model 303 corresponding to the three-channel input that has been trained using a large amount of image data other than the composite image. In other words, the foreign body inspection device 1 can perform image inspection (foreign body inspection) using the deep learning model 303 corresponding to the three-channel input, using the composite image IMG_CM as an input.

[0075] Therefore, even when multiple light sources are used to capture an image of the target object TG, the foreign body inspection device 1 does not need to construct a new deep learning model that supports input of four or more channels. In other words, the foreign body inspection device 1 can reduce the cost of constructing a new deep learning model.

[0076] Furthermore, the foreign matter inspection device 1 can perform transfer learning using a relatively small amount of learning data (synthetic image) for a deep learning model that has already been trained using image data other than the synthetic image. Therefore, the foreign matter inspection device 1 can reduce the cost required for collecting learning data.

[0077] 2. Second embodiment A foreign matter inspection device according to a second embodiment will be described. A foreign matter inspection device 1B according to the second embodiment differs from the first embodiment in that it includes a display unit 14. Also, in the foreign matter inspection device 1B according to the second embodiment, a processing unit 11B differs from the first embodiment. In the following description, the differences from the first embodiment will be described.

[0078] 2.1 Configuration of the foreign body inspection device The configuration of a foreign matter inspection apparatus 1B according to this embodiment will be described with reference to Fig. 10. Fig. 10 is a block diagram showing an example of the configuration of the foreign matter inspection apparatus 1B.

[0079] 10, the foreign matter inspection device 1B includes a processing unit 11B, a light source unit 12, an imaging unit 13, and a display unit 14. The configuration other than the processing unit 11B and the display unit 14 is similar to the configuration in FIG. 1 shown in the first embodiment.

[0080] In addition to the operation control process and the foreign object determination process, the processing unit 11B transmits to the display unit 14 a determination result (determination information) obtained by the foreign object determination process.

[0081] The display unit 14 is, for example, a display, and displays the determination result (determination information) received from the processing unit 11.

[0082] 2.2 Hardware configuration of the foreign body inspection device The hardware configuration of the foreign matter inspection apparatus 1B will be described with reference to Fig. 11. Fig. 11 is a block diagram showing an example of the hardware configuration of the foreign matter inspection apparatus 1B.

[0083] As shown in FIG. 11, the processing unit 11B includes a processor 111, a storage medium 112, a communication interface 113B, and a user interface 114.

[0084] The communication interface 113B is responsible for communication between the light source unit 12, the photographing unit 13, the display unit 14, and other external devices or apparatuses and the processing unit 11B. The communication interface 113B transmits control commands to the light source unit 12 and the photographing unit 13, for example. The communication interface 113B receives captured images from the photographing unit 13, for example. The communication interface 113B also transmits, for example, a determination result (determination information) obtained by the foreign object determination process to the display unit 14 and other external devices or apparatuses.

[0085] The hardware configuration other than the display unit 14 and the communication interface 113B is similar to the hardware configuration shown in FIG. 2 in the first embodiment.

[0086] 2.3 Functional configuration of the processing section The functional configuration of the processing unit 11B will be described with reference to Fig. 12. Fig. 12 is a block diagram showing an example of the functional configuration of the processing unit 11B. Fig. 12 also shows the light source unit 12, the photographing unit 13, and the display unit 14.

[0087] 12, the processing unit 11B includes a process execution unit 211B and a storage unit 212. The process execution unit 211B includes an operation control unit 221 and a determination processing unit 222B.

[0088] The determination processing unit 222B determines the presence or absence of cartilage in the dressed meat TG1. Furthermore, the determination processing unit 222B transmits the determination result (determination information) obtained by the foreign matter determination process to the display unit .

[0089] The configuration of the functional blocks of the determination processing unit 222B is similar to the configuration of the functional blocks of the determination processing unit 222 shown in FIG. 5 in the first embodiment.

[0090] 2.4 Effects of this embodiment With the configuration according to this embodiment, the same effects as those of the first embodiment can be obtained.

[0091] 3. Third embodiment A foreign matter inspection device according to a third embodiment will be described. In the foreign matter inspection device 1C according to the third embodiment, a light source unit 12C and an imaging unit 13C are different from those in the first embodiment. Furthermore, the foreign matter inspection device 1C according to the third embodiment differs from the first embodiment in that it includes a light-shielding box 15. In the following description, the differences from the first embodiment will be described.

[0092] 3.1 Hardware configuration of the foreign body inspection system The hardware configuration of the foreign matter inspection apparatus 1C will be described with reference to Fig. 13. Fig. 13 is a block diagram showing an example of the hardware configuration of the foreign matter inspection apparatus 1C.

[0093] The foreign matter inspection device 1C includes a processing section 11, a light source section 12C, an imaging section 13C, and a light-shielding box 15.

[0094] The light source unit 12C includes an ultraviolet light source 121C, a visible light source 122C, and an infrared light source 123C. The ultraviolet light source 121C includes a plurality of ultraviolet LEDs. The visible light source 122C includes a plurality of visible light LEDs. The infrared light source 123C includes a plurality of infrared LEDs. The plurality of ultraviolet LEDs in the ultraviolet light source 121C are arranged in a ring shape when viewed from the transport unit CV side (target TG side) as in FIG. 3 shown in the first embodiment. The plurality of visible light LEDs in the visible light source 122C and the plurality of infrared LEDs in the infrared light source 123C are also arranged in a ring shape when viewed from the transport unit CV side as in FIG. 3 shown in the first embodiment. That is, the ultraviolet light source 121C, the visible light source 122C, and the infrared light source 123C are arranged in a concentric shape with respect to each other.

[0095] The photographing unit 13C includes a camera 134. The camera 134 is, for example, a camera sensitive to ultraviolet light, visible light, and infrared light. The camera 134 is disposed so as to be surrounded by a plurality of ultraviolet LEDs in the ultraviolet light source 121C when viewed from the transport unit CV side.

[0096] In the example of FIG. 13, the ultraviolet light source 121C, the visible light source 122C, and the infrared light source 123C are arranged in the order of ultraviolet light source 121C, visible light source 122C, and infrared light source 123C from the camera 134 side, but the arrangement order of the ultraviolet light source 121C, the visible light source 122C, and the infrared light source 123C is not limited to this.

[0097] 13, the ultraviolet LEDs in the ultraviolet light source 121C are arranged in a ring shape when viewed from the transport section CV side, but the arrangement of the ultraviolet LEDs in the ultraviolet light source 121C is not limited to this. For example, the ultraviolet LEDs in the ultraviolet light source 121C may be arranged side by side in the transport direction of the transport section CV. The same applies to the visible LEDs in the visible light source 122C and the infrared LEDs in the infrared light source 123C.

[0098] The light-shielding box 15 is a member for blocking ambient light. The upper end of the light-shielding box is fixed to a support part SP. The light-shielding box 15 has, for example, an entrance (not shown) for carrying in the fresh meat TG1, and an exit (not shown) for carrying out the fresh meat TG1. The entrance and the exit can be opened and closed, for example.

[0099] As described above, the camera 134 is sensitive to ultraviolet light, visible light, and infrared light. Therefore, if ambient light is irradiated onto the fresh meat TG1, this may adversely affect the accuracy of images taken in each of the states where the fresh meat TG1 is irradiated with ultraviolet light by the ultraviolet light source 121C, where the fresh meat TG1 is irradiated with visible light by the visible light source 122C, and where the fresh meat TG1 is irradiated with infrared light by the infrared light source 123C. Therefore, in this embodiment, a light-shielding box 15 is provided to prevent the fresh meat TG1 from being irradiated with ambient light during inspection.

[0100] The hardware configuration other than the light source unit 12C, the photographing unit 13C, and the light-shielding box 15 is similar to the hardware configuration shown in FIG. 2 in the first embodiment.

[0101] In one example, when the fresh meat TG1 is carried into the light-shielding box 15, the transport of the fresh meat TG1 from the upstream side to the downstream side in the transport direction of the transport unit CV is temporarily stopped. Then, in a state where the movement of the fresh meat TG1 is stopped, the operation control unit 221 causes the light source unit 12C to sequentially irradiate the fresh meat TG1 with ultraviolet light, visible light, and infrared light. Then, in each of the states where the ultraviolet light is irradiated, the visible light is irradiated, and the infrared light is irradiated, the operation control unit 221 causes the photographing unit 13C to photograph the fresh meat TG1. When the photographing by the photographing unit 13C is finished, the transport of the fresh meat TG1 downstream in the transport direction of the transport unit CV is resumed.

[0102] 3.2 Effects of this embodiment The configuration according to this embodiment can provide the same effects as those of the first embodiment. Note that the third embodiment can be applied to the second embodiment.

[0103] 4. Modifications, etc. The embodiment is not limited to the above-described embodiment, and various modifications are possible.

[0104] For example, in the above embodiment, the light source unit 12 emits ultraviolet light, visible light, and infrared light, but the combination of the three lights is not limited to this. For example, it may be a combination of two ultraviolet lights with different wavelengths and visible light, or a combination of lights with other wavelengths.

[0105] For example, although the above embodiment has been described with respect to a case where there are three light sources, the number of light sources may be two. For example, when the foreign body inspection device 1 captures an ultraviolet light image and a visible light image, an extracted image of one channel may be extracted from the ultraviolet light image, and an extracted image of two channels may be extracted from the visible light image.

[0106] Furthermore, in the flowcharts described in the above embodiments, the order of the processes can be changed as much as possible.

[0107] Although some embodiments of the present invention have been described, these embodiments are presented as examples and are not intended to limit the scope of the invention. These novel embodiments can be implemented in various other forms, and various omissions, substitutions, and modifications can be made without departing from the spirit of the invention. These embodiments and their modifications are included in the scope and spirit of the invention, and are included in the scope of the invention and its equivalents described in the claims. [Explanation of symbols]

[0108] 1, 1B, 1C...foreign substance inspection device, 11, 11B...processing section, 12, 12C...light source section, 13, 13C...imaging section, 14...display section, 15...light-shielding box, 111...processor, 112...storage medium, 113, 113B...communication interface, 114...user interface, 121, 121C...ultraviolet light source, 122, 122C...visible light source, 123, 123C...infrared light source, 131...ultraviolet camera, 132...visible light camera, 133...infrared camera , 134...camera, 211, 211B...processing execution unit, 212...memory unit, 221...operation control unit, 222, 222B...judgment processing unit, 231...image acquisition unit, 232...image extraction unit, 233...image synthesis unit, 234...foreign object judgment unit, 235...judgment information generation unit, 236...learning data acquisition unit, 237...learning processing unit, 301...inspection data, 302...learning data, 303...deep learning model, SP...support unit, CV...transport unit, TG...target object, TG1...meat

Claims

1. an image acquisition unit that acquires a first image captured in a state where an object is irradiated with ultraviolet light, a second image captured in a state where the object is irradiated with visible light, and a third image captured in a state where the object is irradiated with infrared light; an extracting unit that extracts a first extracted image of one channel from the first image, extracts a second extracted image of one channel from the second image, and extracts a third extracted image of one channel from the third image; a synthesis unit that synthesizes the first extracted image, the second extracted image, and the third extracted image to generate a synthesis image; a determination unit that determines a foreign object in the object by using the composite image; Equipped with Foreign object inspection equipment.

2. The determination unit is configured to perform the determination of the foreign object by using a deep learning model that uses the composite image as an input.

2. The foreign matter inspection device according to claim 1.

3. the first extracted image is a B channel image of the first image, the second extracted image is an R channel image of the second image, the third extracted image is a grayscale image of the third image; 2. The foreign matter inspection device according to claim 1.

4. a light source unit configured to be able to irradiate the ultraviolet light, the visible light, and the infrared light to the object; An imaging unit that images the object; Further comprising:

2. The foreign matter inspection device according to claim 1.

5. Obtaining a first image captured while irradiating an object with ultraviolet light, a second image captured while irradiating the object with visible light, and a third image captured while irradiating the object with infrared light; extracting a first extracted image of one channel from the first image; extracting a second extracted image of one channel from the second image; extracting a third extracted image of one channel from the third image; combining the first extracted image, the second extracted image, and the third extracted image to generate a three-channel composite image; determining a foreign object in the object using the composite image; and Equipped with Foreign body inspection method.

6. The determining step uses a deep learning model that uses the synthetic image as an input. The method for inspecting for foreign matter according to claim 5.

Images