JP7734912B2 - Foreign object detection device and foreign object detection method - Google Patents

Description

The present invention relates to a foreign object detection device and a foreign object detection method.

In food production, contamination with foreign objects can be a serious problem. Therefore, to prevent food contaminated with foreign objects from being shipped, food manufacturers typically use foreign object detection equipment to inspect food. X-ray inspection machines and metal detectors, in particular, are often used in food production because they can detect foreign objects without destroying the food.

For example, Patent Document 1 discloses an X-ray inspection device that includes a housing having an opening on each side that serves as the entrance or exit for the food under test, and a pair of side covers each having an inclined surface and two side panels for covering the opening; an endless belt conveyor that transports the food under test from the entrance opening to the exit opening; an X-ray source within the housing that irradiates the food under test on the belt conveyor with X-rays; and X-ray detection means that detects the X-rays that have passed through the food under test and detects any foreign matter mixed in the food under test, wherein the endless belt conveyor is an isosceles trapezoidal cross section that has an upward slope, an upper horizontal surface, a downward slope, and a lower horizontal surface, and the length and slope of the upward slope and downward slope in the transport direction are configured to be approximately equal to the length and slope of the two inclined surfaces of the side covers.

Furthermore, foreign object detection devices that detect foreign objects by irradiating them with illumination light may also be used. For example, Patent Document 2 discloses a food foreign object detection device that includes a transport means for transporting food to a foreign object detection area, an imaging means for capturing a surface image of the food from one side of the detection area, an illumination means for irradiating the food with illumination light from one or both sides of the inspection area across the transport means, and an image processing means for performing image processing, including binarization, on the captured surface image.

JP 2017-075830 A Special Publication No. 2008-541007

However, foreign objects such as bones, shells, seeds, stems, roots, and soft fibrous materials that can be mixed into solid foods such as meat, seafood, fruit pulp, and potatoes often have a density that is only slightly different from that of the solid food itself, and they usually do not contain metallic substances, making them difficult to detect with X-ray inspection machines or metal detectors. Detecting such foreign objects buried inside solid foods is particularly difficult. Furthermore, as described in Patent Document 1, X-ray inspection machines must minimize the leakage of X-rays to the outside, which poses safety and cost issues.

Furthermore, the foreign object detection device described in Patent Document 2 essentially irradiates food with illumination light from both above and below to detect foreign objects such as hair or insects attached to the surface of the food, but this also poses the problem of difficulty in detecting foreign objects such as those mentioned above that are buried inside solid food.

The present invention therefore aims to provide a foreign object detection device and method capable of detecting foreign objects mixed into solid food.

In order to solve the above problem, the inventors conducted extensive research and discovered that the above-mentioned foreign matter has a lower near-infrared light transmittance than the solid food that contains it. Furthermore, they discovered that an effective way to easily detect such foreign matter is to locally irradiate the solid food with near-infrared light, allow it to pass through, capture an image of the solid food that has locally transmitted near-infrared light, and use the captured image to detect foreign matter, leading to the completion of the present invention.

That is, the present invention provides the following (1) to (11).
(1) A foreign object detection device that detects foreign objects mixed into solid food and having a lower near-infrared light transmittance than the solid food, comprising a near-infrared light irradiation unit that locally irradiates the solid food with near-infrared light, an imaging unit that captures an image of the solid food that has transmitted the near-infrared light, and a detection unit that detects the foreign object from the captured image.
(2) The foreign object detection device described in (1), wherein the near-infrared light irradiation unit includes a light source that emits the near-infrared light, and a near-infrared light adjustment unit that adjusts the amount and/or direction of the near-infrared light that is emitted from the light source and passes through the solid food.
(3) A foreign object detection device as described in (2), wherein the near-infrared light adjustment unit is arranged between the light source and the solid food and includes a slit that narrows a portion of the light direction of the near-infrared light emitted from the light source.
(4) The foreign object detecting device according to (3), wherein the slit has a width in the short direction of 10 mm or less and a length in the long direction of the slit that is equal to or greater than the dimension of the solid food in the long direction.
(5) A foreign object detection device described in any one of (1) to (4), which is provided with a compression and fixing means that applies a compression load to the solid food that is smaller than a crushing compression load that is the compression load at which the solid food is crushed, thereby compressing and fixing the thickness of the solid food in the direction that transmits the near-infrared light, and the imaging unit captures the image of the compressed solid food by transmitting the near-infrared light.
(6) The foreign object detecting device according to (5), wherein the compressing and fixing means compresses and fixes the solid food to a thickness of 30% or more and 90% or less.
(7) The foreign object detecting device according to any one of (1) to (6), wherein the near-infrared light has a wavelength of 800 nm or more and 1200 nm or less.
(8) A foreign object detection device described in any one of (1) to (7), wherein the detection unit detects the foreign object from the images captured by the imaging unit using a machine learning model trained by providing data of defective images of the solid food containing the foreign object and good images of the solid food not containing the foreign object, or a machine learning model trained by providing data of the good images.
(9) A foreign object detection device described in any one of (1) to (8), wherein the solid food is meat, seafood, fruit pulp, or potatoes, and the foreign object is one or more selected from the group consisting of bones, shells, seeds, and soft foreign objects.
(10) A method for detecting foreign objects mixed into solid food and having a lower near-infrared light transmittance than the solid food, comprising a near-infrared light irradiation process for locally irradiating the solid food with near-infrared light, an imaging process for capturing an image of the solid food through which the near-infrared light has passed, and a foreign object detection process for detecting the foreign object from the captured image.
(11) A method for detecting a foreign object from a solid food product, comprising: a near-infrared light irradiation step of locally irradiating a solid food product with near-infrared light; an imaging step of capturing an image of the solid food product through which the near-infrared light has passed; a foreign object detection step of detecting a foreign object having a lower near-infrared light transmittance than the solid food product from the captured image; and a foreign object removal step of removing the detected foreign object from the solid food product or removing the solid food product containing the foreign object.
Methods for producing solid foods.

The present invention provides a foreign object detection device and a foreign object detection method that can easily detect foreign objects that have a lower near-infrared light transmittance than solid foods and that have been mixed in with the solid foods. This device or method can then easily detect the foreign objects that have been mixed in with solid foods.

1 is a schematic diagram of a foreign object detection device according to an embodiment of the present invention. FIG. 2 is a front view of a near-infrared light adjusting section including a slit of the foreign object detecting device according to the embodiment. 1 is an image of scallop meat (adductor muscle) photographed in Example 1 (photograph used as a substitute for a drawing). 1 is an image of chicken taken in Example 2 (a photograph as a substitute for a drawing). 1 is an image of prune flesh photographed in Example 3 (photograph substitute for drawing). 1 is an image of scallop meat (adductor muscle) photographed in Example 4 (photograph substitute for drawing). 10 is an image of a potato taken in Example 5 (a photograph as a substitute for a drawing).

The present invention will now be described.
The present invention relates to a foreign object detection device (hereinafter also referred to as the "foreign object detection device of the present invention") that includes a near-infrared light irradiation unit that locally irradiates near-infrared light onto solid food, an imaging unit that captures an image of the solid food that has transmitted the near-infrared light, and a detection unit that detects foreign objects that have a lower near-infrared light transmittance than the solid food from the captured image.The present invention also includes a foreign object detection method (hereinafter also referred to as the "foreign object detection method of the present invention") that uses the foreign object detection device of the present invention and includes a near-infrared light irradiation step that locally irradiates near-infrared light onto solid food, an imaging step that captures an image of the solid food that has transmitted the near-infrared light, and a foreign object detection step that detects foreign objects from the captured image, as well as a solid food manufacturing method (hereinafter also referred to as the "solid food manufacturing method of the present invention") that further includes a foreign object removal step that removes foreign objects detected by this method or the like from the solid food or removes solid food containing foreign objects.

First, an embodiment of a foreign object detection device according to the present invention will be described using Figures 1 and 2. Note that in these drawings, similar components are given the same reference numerals, and duplicate descriptions will be omitted where appropriate. Also, for convenience, some parts are not given reference numerals (omitted). Furthermore, the dimensional proportions of the devices and components shown in the drawings may differ from their actual dimensional proportions in order to facilitate understanding of the invention.

<Overall configuration of foreign object detection device>
The foreign object detector 100 according to the present invention is a device for detecting foreign objects 13 that are mixed into solid food 11 (especially those that are partially or completely buried inside) and have a lower near-infrared light transmittance than the solid food 11. Furthermore, the detection of such foreign objects 13 can also be performed in-line.

The target solid food 11 is a food that is solid at room temperature (10°C to 35°C), and typical examples include meat (chicken, pork, beef, etc.), seafood (fish meat, cephalopod meat such as squid and octopus, shelled seafood such as scallops, sea urchins, shrimp, and crabs), fruit pulp (edible fruit pulp such as prunes and berries, which may include the edible skin), and potatoes (potatoes, sweet potatoes, taro, etc.). These may be fresh (including frozen and thawed) or processed (boiled, blanched, dried, etc.). The foreign object detection device 100 of the present invention is particularly suited to soft solid foods, i.e., solid foods 11 that can be compressed without crushing to between 30% and 90% of their thickness in at least one direction by applying a compressive load smaller than the crushing load, as described below. All of the above examples are soft solid foods that can be compressed without crushing to a thickness of 30% to 90% in at least one direction. However, solid food 11 does not include foods that are liquid or paste-like at room temperature. The foreign object detection device 100 of the present invention is extremely effective in detecting bones in livestock meat, bones and scales in fish meat, shells in the flesh (meat, etc.) of shelled seafood, seeds in fruit pulp, or soft foreign objects in potatoes.

Furthermore, foreign matter 13 having a lower near-infrared light transmittance than the target solid food 11 is less susceptible to near-infrared light 23 than the solid food 11 and is undesirable or should be prevented from being mixed in when the solid food 11 is made into a finished product. Examples of foreign matter include bones (including small bones and cartilage), scales, shells (including seashells and sea urchin spines), seeds (hard seeds, etc.), and soft foreign matter (stems (the stems connecting the fruit to the branches), roots, stalks, soft fibrous materials used in manufacturing processes, etc.), and some of these (for example, bone fragments, scale fragments, shell fragments, seed fragments, stem fragments, etc.) are also included. These substances are difficult to detect with X-ray machines or metal detectors because they are primarily derived from living organisms and often have a small density difference with the target solid food 11, and are non-metallic. Detection is particularly difficult when they are partially or completely buried inside the solid food 11. However, because they often have a lower near-infrared light transmittance than the solid food 11 and the difference is large, the foreign object detection device 100 of the present invention is characterized by being able to easily detect one or more selected from the group consisting of these substances that have become mixed into the solid food 11. In addition to these, any substance that has a lower near-infrared light transmittance than the target solid food 11 and whose inclusion in the solid food 11 is undesirable or should be prevented when the solid food 11 is made into a finished product can be detected as a foreign object 13.

The foreign object detection device 100 according to the present invention comprises at least a near-infrared light emitting unit 20, an imaging unit 41, and a detection unit 51 (see, for example, Figure 1). In addition, as shown in Figure 1, an embodiment may also comprise a control unit 61, a compression and fixing unit 71, a conveying unit 73, etc. These will be described in detail below.

<Near infrared light irradiation section>
The near-infrared light irradiating unit 20 is a device capable of locally irradiating the solid food 11 with near-infrared light 23. Therefore, it includes at least a light source 21 that emits the near-infrared light 23.
Here, "locally irradiating" near-infrared light 23 includes not only irradiating only a portion of solid food 11 with near-infrared light 23, but also irradiating the entire solid food 11 with weak near-infrared light 23 and irradiating a portion with stronger near-infrared light 23. Furthermore, "near-infrared light" refers to infrared light (light with a wavelength longer than that of visible light) with a wavelength of 780 nm or more and 2500 nm or less, which is closer to visible light. Light source 21 is not limited as long as it is capable of emitting such near-infrared light, but may include an LED light (e.g., UPD2450W-2S manufactured by U-Technology Co., Ltd.) or a halogen light (e.g., LA-100IR manufactured by Hayashi Repic Co., Ltd.).

Examples of embodiments for locally irradiating near-infrared light 23 include an embodiment in which the light source 21 itself locally emits near-infrared light 23, an embodiment in which the near-infrared light 23 emitted from the light source 21 is partially blocked by a shielding object, or an embodiment in which the light direction is partially focused or aligned by a mirror or polarizing plate. This makes it possible to suppress scattering of near-infrared light 23 inside the solid food 11, making it easier to capture a clear image of the foreign object 13 even if it is buried inside the solid food 11, thereby making it easier to detect the foreign object 13. Furthermore, in the foreign object detection device 100 according to the present invention, the near-infrared light irradiator 20 is preferably an embodiment that includes a light source 21 that emits near-infrared light 23 and a near-infrared light adjuster 31 (such as the aforementioned shield, mirror, or polarizing plate) that adjusts the amount and/or direction of the near-infrared light 23 that is emitted from the light source 21 and passes through the solid food 11, as this makes the device easier to design, allows the device to be obtained at low cost, and also makes it easier to detect foreign objects 13.

In particular, it is more preferable that the near-infrared light adjustment unit 31 includes a slit 31a, as shown in FIG. 2, disposed between the light source 21 and the solid food 11 as a shield that blocks a portion of the near-infrared light 23, and that narrows a portion of the light direction of the near-infrared light 23 emitted from the light source 21. In other words, it is more preferable that the near-infrared light 23 (radiated light 23a) emitted from the light source 21 is narrowed in at least a portion of its light direction by passing through the slit 31a of the near-infrared light adjustment unit 31, and that the near-infrared light 23 that passes through the slit 31a is transmitted through the solid food 11 (transmitted light 23b). In other words, it is more preferable that the near-infrared light adjustment unit 31 includes a member having a slit 31a formed therein, narrowing a portion of the light direction of the near-infrared light 23 emitted from the light source 21. Regardless of the type of light source 21, scattering of near-infrared light 23 inside the solid food 11 is easily suppressed, and therefore even if the light intensity is set to a higher level, the image is less likely to turn white and the foreign object 13 becomes indistinguishable, making it easier to detect the foreign object 13.

For example, the near-infrared light adjustment unit 31 may be a plate-like structure or other member made of a material capable of blocking near-infrared light 23, with a rectangular cutout slit 31a formed in part of it (Figure 2). Examples of materials capable of blocking near-infrared light 23 include metal, wood, colored resin, and paper. The shape of the slit 31a is not limited to a rectangular shape, and may be an elliptical shape, for example.

From the viewpoint of suppressing scattering of near-infrared light 23, it is preferable that the width D (maximum width) of the slit 31a in the short side direction is 10 mm or less, more preferably 8 mm or less, and even more preferably 5 mm or less, and that the length L in the long side direction is equal to or greater than the dimension of the solid food 11 in the long side direction. Furthermore, from the viewpoint of easily obtaining the amount of light necessary to detect foreign matter 13, it is more preferable that the minimum width D of the slit 31a is 0.5 mm or more, and even more preferably 1 mm or more.

Furthermore, although not limited thereto, the configuration of the slit 31a described above may be such that when both the near-infrared light irradiator 20 and the solid food 11 are stationary, the surface area of the region on the solid food 11 that is locally irradiated with the near-infrared light 23 is preferably 20% or less, more preferably 17% or less, even more preferably 10% or less, and even more preferably 5% or less of the total surface area of the solid food 11.

When the near-infrared light adjusting unit 31 includes the above-mentioned slit 31a, in order to more reliably detect foreign matter 13 mixed in the solid food 11, it is preferable to move the solid food 11 or the slit 31a so that the slit 31a moves in a constant direction and at a constant speed relative to the solid food 11, and continuously capture images of the solid food 11 during this movement. It is then preferable to combine these continuously captured images and use them as an image for detecting foreign matter 13.

Note that if the light source 21 itself is configured to be capable of locally emitting near-infrared light 23 as described above (such as a light source 21 capable of emitting near-infrared laser light), the near-infrared light irradiator 20 may not include the near-infrared light adjuster 31 described above.

Furthermore, since it is easier to detect the types of foreign matter 13 described above, it is preferable that the near-infrared light 23 locally irradiated onto the solid food 11 from the near-infrared light irradiating unit 20, i.e., the near-infrared light 23 that passes through the solid food 11, has a wavelength of 800 nm or more and 1200 nm or less. The lower limit of this wavelength is more preferably 900 nm or more, and even more preferably 950 nm or more. The upper limit of this wavelength is also more preferably 1100 nm or less, and even more preferably 1050 nm or less.

<Image capture unit>
The imaging unit 41 is a device capable of capturing an image of the solid food 11 through which the near-infrared light 23 has passed (an image of the solid food 11 including the transmitted light 23b), and is exemplified by an image capturing camera, for example. The imaging unit 41 may be configured to capture the above-described image, and its location is not particularly limited, for example, in a configuration including a mirror together with the image capturing camera. However, from the viewpoint of ease of designing the device, it is preferable that the imaging unit 41 be located opposite the near-infrared light irradiator 20 with respect to the solid food 11.

The image capturing camera is not limited to, but preferably includes, a camera with high sensitivity in the near-infrared wavelength range, such as an InGaAs camera equipped with an InGaAs (indium gallium arsenide) sensor; a black silicon camera, which is cheaper than other cameras and can also capture images in the visible light range; or a hyperspectral camera capable of scanning multiple wavelengths. The camera's exposure time (shutter speed) when capturing an image can be set based on the amount of near-infrared light 23 transmitted through the solid food 11, and is, for example, not limited to, approximately 1,000 μs or more and 150,000 μs or less.

<Detection unit>
The detection unit 51 is a device capable of detecting the foreign object 13 from the image captured by the imaging unit 41, and is preferably a computer. Therefore, for example, the detection unit 51 may be a computer including a storage device 51a and a processing unit 51b connected via a network or the like. If necessary, the detection unit 51 may also include an input unit such as a keyboard, mouse, or touch panel for inputting settings, and a monitor for displaying captured images. From the perspective of improving the detection accuracy of the foreign object 13, the detection unit 51 is preferably connected to a control unit 61 (e.g., a control computer) that controls at least some of the devices included in the foreign object detection device 100 according to the present invention, such as the light source 21 and the imaging unit 41, and is controlled in conjunction with the emission of near-infrared light 23 from the light source 21 and the timing of imaging by the imaging unit 41.

The detection unit 51 reads the image captured by the imaging unit 41 into a storage device 51a or the like, and, if necessary, combines multiple images to create image data. The processor 51b or the like then uses this image data to determine the presence or absence of foreign matter 13 based on the intensity (brightness) of the transmitted light 23b, and if foreign matter 13 is present, identifies its location. The results are then output to a monitor or other device.

Furthermore, the detection unit 51 (e.g., the arithmetic processing device 51b) is preferably configured to detect foreign matter 13 by using a machine learning model trained on defective images of solid food 11 containing foreign matter 13 and non-defective images of solid food 11 not containing foreign matter 13 as data (trainer data) from images captured by the imaging unit 41 to determine whether foreign matter 13 is present or absent, or by using a machine learning model trained on only non-defective images of these as data (trainer data) to determine that solid food 11 containing foreign matter 13 is a defective solid food. In other words, it is preferable to detect foreign matter 13 using a machine learning model trained on at least non-defective images as data. This is because this can speed up foreign matter detection and also improves the accuracy of foreign matter detection.

This machine learning model (ML model), also known as a trained model or an AI (Artificial Intelligence) model, is a model obtained by machine learning using training data, i.e., supervised learning. For example, but not limited to, the ML model may be a regression equation obtained by regression analysis, or may be configured with a convolutional neural network including a deep neural network, and may be realized by a combination of a computer program and parameters, a combination of multiple functions and parameters, or the like.
In addition, a method other than the machine learning model may be used, such as setting a brightness threshold for identifying foreign matter 13 in an image in advance and storing it in the memory means of the detection unit 51, and the detection unit 51 determining whether or not a foreign matter 13 is present by comparing this threshold with the threshold.

<Compression fixing means>
Furthermore, the foreign object detection device 100 according to the present invention preferably includes a compression and fixation means 71 that applies a compression load to the solid food 11 that is smaller than the crushing compression load that crushes the solid food 11, thereby compressing and fixing the thickness of the solid food 11 in the direction through which near-infrared light 23 is transmitted (or in at least one direction if there are multiple transmission directions). In this case, the imaging unit 41 captures an image of the solid food 11 compressed by the compression and fixation means 71 and transmitting near-infrared light 23. Therefore, in this configuration, it is important to fix the solid food 11 using a tool or the like so that the thickness does not return to its original state. It is also important to compress the solid food 11 to a degree that will not cause it to fracture and to maintain that state. With this configuration, scattering of the transmitted near-infrared light 23 (particularly scattering of near-infrared light 23 at the periphery of the foreign object 13 that does not directly hit the foreign object 13) can be further suppressed due to the effects of reducing the distance between the foreign object 13 inside the solid food 11 and the surface of the solid food 11 in the direction through which the near-infrared light 23 passes and reducing the amount of unevenness on the surface of the solid food 11.On the other hand, the near-infrared light transmittance of areas of the solid food 11 that do not contain the foreign object 13 is further increased, thereby making it possible to obtain a clearer image of the foreign object 13 and making it easier to detect the foreign object 13.

The compression and fixation means 71 is preferably configured to compress and fix the thickness of the solid food 11 in the direction through which near-infrared light 23 is transmitted to between 30% and 90% by a compression load smaller than the crushing compression load, from the viewpoint of both maintaining the shape of the solid food 11 and facilitating detection of foreign matter 13. Therefore, in a configuration including such a compression and fixation means 71, it is preferable to use the aforementioned soft solid foods that can be compressed and fixed without crushing to between 30% and 90% of the thickness in the direction through which near-infrared light 23 is transmitted. The lower limit of this compression is preferably 33% or more, more preferably 40% or more, and even more preferably 50% or more. The upper limit is more preferably 85% or less, even more preferably 80% or less, and even more preferably 75% or less. For example, in the case of prune pulp, it is preferable to compress a thickness of approximately 16 to 25 mm to a thickness of approximately 10 to 15 mm, within the above range. Here, the compression percentage indicates the thickness after compression as a percentage, assuming that the thickness of the solid food 11 in the direction through which the near-infrared light 23 is transmitted before compression is 100%.

The specific configuration of this compression and fixing means 71 is not limited to, but may be exemplified by an embodiment in which the solid food 11 is fixed by being sandwiched between plate-shaped members (such as plates made of acrylic resin or glass) that are transmissive to near-infrared light 23, in the direction that transmits the near-infrared light 23.

<Transportation means>
Furthermore, the foreign object detection device 100 according to the present invention preferably includes a conveying means 73 capable of conveying the target solid food 11 (e.g., soft solid food compressed and fixed by the compressing and fixing means 71) to a process in which near-infrared light 23 is transmitted and then conveying the food from this process to the next process. This conveying means 73 may be, for example, a belt conveyor with a drive mechanism as shown in FIG. 1 , but must be configured so as not to interfere with the irradiation of the conveyed solid food 11 with near-infrared light 23 and the imaging thereof (e.g., a conveyor that is transparent to near-infrared light, a conveyor with a gap in the near-infrared light irradiation area, etc.). Furthermore, this conveying means 73 is preferably connected to the control unit 61, so that the conveying speed, etc., can be controlled.

When the foreign object detection device 100 according to the present invention is equipped with such a conveying means 73, it is preferable that, as shown in FIG. 1, the near-infrared light irradiator 20 (the light source 21 and the near-infrared light adjuster 31 including the slit 31a) and the image capturer 41 are fixedly positioned, and the conveying means 73 moves the solid food 11 at a constant conveying speed while the image capturer 41 captures images at predetermined time intervals. Note that the direction of movement (conveyance direction) in this case coincides with the width direction of the slit 31a, as indicated by the black arrow in FIG. 1, for example. With this configuration, images of the solid food 11 that have been transmitted through the near-infrared light 23 can be continuously captured, and the resulting images can be processed by the detection unit 51, facilitating detection of foreign objects 13 regardless of their position in the solid food 11.

Furthermore, in this configuration, if the near-infrared light irradiation unit 20 includes a slit 31a as the near-infrared light adjustment unit 31, it is preferable that the distance the solid food 11 is conveyed by the conveying means 73 during the imaging time interval, i.e., the length of movement of the slit 31a for each captured image, is smaller than the width D of the slit 31a. This is because at least one image of every portion of the solid food 11 can be captured as it passes through the slit 31a. Note that when the conveying means 73 moves the solid food 11, the solid food 11 may be moved continuously without stopping, or may be moved intermittently by repeatedly moving and stopping. When moving intermittently, it is preferable that the control unit 61 control the operation of the conveying means 73 and the imaging unit 41 so that the imaging unit 41 captures images while the solid food 11 is stopped. This allows the imaging unit 41 to capture images with minimal blur. Furthermore, when the conveying means 73 moves the solid food 11 intermittently, it is preferable that the length of each movement (conveyance distance) be smaller than the width D of the slit 31a, as described above.

In the above embodiment, the image capturing unit 41 captures images while the transport means 73 transports the solid food 11, but this is not limited to this. For example, the light source 21 may be provided with a drive mechanism (not shown) that changes the irradiation direction of the near-infrared light 23, and the image capturing unit 41 may capture images while sequentially changing the position at which the near-infrared light 23 is irradiated onto the stationary solid food 11, or the image capturing unit 41 may capture images while sequentially changing the position of the near-infrared light adjusting unit 31 (e.g., a member including a slit 31a) between the fixedly positioned light source 21 and the stationary solid food 11.

Furthermore, the foreign object detection device 100 according to the present invention may include any other equipment or components as long as they do not affect the effects of the present invention. For example, it may be equipped with a rejection mechanism (air, rejection conveyor, etc.) that rejects solid food 11 containing foreign objects 13 based on the captured image, or auxiliary lighting other than the light source 21 that can improve the accuracy of image capture.

<Foreign object detection method, etc.>
Next, an embodiment of a foreign object detection method and the like according to the present invention will be described.

The foreign object detection method according to the present invention can be implemented using the foreign object detection device 100 according to the present invention, and includes a near-infrared light irradiation step of locally irradiating near-infrared light 23 onto solid food 11, an imaging step of capturing an image of solid food 11 through which this near-infrared light 23 has passed, and a foreign object detection step of detecting foreign objects 13 from the captured image. The details of each step may be the same as those of the foreign object detection device 100 according to the present invention described above.

Furthermore, the foreign object detection method of the present invention may include a compressing and fixing step using a compressing and fixing means that applies a compressive load smaller than the crushing compressive load to the solid food 11, thereby compressing and fixing the thickness of the solid food 11 in the direction through which the near-infrared light 23 is transmitted. It may also include a conveying step in which the solid food 11 is conveyed by a conveying means. These steps may be similar in configuration to the foreign object detection device 100 of the present invention described above. For example, the method may involve moving the solid food 11 compressed and fixed by the compressing and fixing means 71 at a constant conveying speed by the conveying means 73 (compressing and fixing step, conveying step), locally irradiating the compressed solid food 11 with near-infrared light 23 using the near-infrared light irradiating unit 20 (near-infrared light irradiation step), capturing images of the solid food 11 through which the near-infrared light 23 has transmitted at predetermined time intervals using the imaging unit 41 (imaging step), and detecting foreign objects 13 using a composite of the captured images (foreign object detection step).

The foreign object detection device 100 or foreign object detection method of the present invention makes it possible to efficiently detect foreign objects 13 such as bones, shells, seeds, stems, and roots that are difficult to detect with X-ray inspection machines or metal detectors, particularly foreign objects 13 buried inside the solid food 11 (which are difficult or impossible to see from the outside). Furthermore, by combining the steps of the foreign object detection method of the present invention with a foreign object removal process in which detected foreign objects 13 are removed from the solid food 11 or solid food 11 containing foreign objects 13 is removed from the manufacturing process, a solid food manufacturing method of the present invention can be provided that can produce solid food 11 that does not contain the above-mentioned foreign objects 13. For example, a solid food manufacturing method can be provided that includes the above-mentioned near-infrared light irradiation process, imaging process, and foreign object detection process, followed by the above-mentioned foreign object removal process. This manufacturing method may further include the above-mentioned compressing and fixing process and conveying process.

Examples of the foreign matter removal process include a process of sucking out foreign matter 13 contained in solid food 11 using a robot or the like, and a rejection process in which solid food 11 containing foreign matter 13 is mechanically rejected (sorted) from the manufacturing process using a defective product rejection device (sorting mechanism, etc.). Furthermore, this foreign matter removal process may also include a process in which foreign matter 13 is manually removed from solid food 11 rejected from the manufacturing process, and the solid food 11 from which foreign matter 13 has been removed is returned to the manufacturing process.

The following describes examples of the present invention, but the present invention is not limited to these examples, and various modifications are possible within the technical concept of the present invention.

Example 1
Boiled oyster adductor muscles (10 mm or 15 mm thick) containing shell fragments as foreign matter were irradiated with near-infrared light at a wavelength of 1050 nm using an LED light (UPD2450W-2S, manufactured by U-Technology Co., Ltd.), and images of the adductor muscles that had been irradiated with near-infrared light were taken using an InGaAs camera (XEVA-165 SS-1700, manufactured by Xenix Corporation, resolution 300 × 256 pixels) under the following conditions.

<a> The adductor muscle was left in its original state (thickness 10 mm), and near-infrared light was irradiated onto the entire muscle so that it penetrated in the thickness direction (no compression, no slits). Images were taken with a camera exposure time of 6000 μs.
<b> The adductor muscle was left in its original state (thickness 10 mm or 15 mm), and near-infrared light was locally irradiated through a 5 mm or 1 mm slit in the short side direction formed in a metal plate so that it could be transmitted in the thickness direction (no compression x 5 mm slit, no compression x 1 mm slit). Images were taken with a camera exposure time of 20,000 μs or 80,000 μs.
<c> An adductor muscle (10 mm thick) was sandwiched between acrylic plates and compressed (compressed without crushing) to a thickness of 6 mm, and fixed in place. Near-infrared light was passed through a 5 mm slit in the short direction formed in the metal plate so that it could be transmitted in the direction of compression (6 mm compression x 5 mm slit). Images were taken with a camera exposure time of 10,000 μs.
<d> An adductor muscle (15 mm thick) was sandwiched between acrylic plates and compressed (compressed without crushing) to a thickness of 12 mm, and fixed in place. Near-infrared light was passed through a 1 mm slit in the short direction formed in a metal plate so that it could be transmitted in the direction of compression (12 mm compression x 1 mm slit). Images were taken with a camera exposure time of 80,000 μs.

The images obtained are shown in Figure 3. The image on the far left of Figure 3 is image <a>, the two images at the top and bottom of the center are images <b>, the top right image is image <c>, and the bottom right image is image <d>. Note that all images under conditions using slits were taken continuously while moving the metal plate on which the slits were formed so that the slits moved from left to right in the image in Figure 3, and these images were then composited (images under conditions using slits in the following examples were also taken in a similar manner unless otherwise noted).

The results show that in <a>, where near-infrared light was irradiated over the entire surface without a slit and allowed to pass through, it was difficult to detect shell fragments, but in <b>, where near-infrared light was irradiated locally with a slit and allowed to pass through, it was easier to detect shell fragments. Furthermore, in <c> and <d>, where the adductor muscle was compressed and fixed to 60% or 80% of its thickness and then irradiated locally with near-infrared light with a slit and allowed to pass through, the shell fragments became clearer and were even easier to detect.

Example 2
Chicken meat (raw chicken fillet, 24 mm or 20 mm thick) containing bone (cartilage or hard bone) as a foreign body was irradiated with near-infrared light at a wavelength of 1050 nm using an LED light (UPD2450W-2S, manufactured by U-Technology Co., Ltd.), and images of the chicken meat that had been irradiated with near-infrared light were taken using an InGaAs camera (XEVA-165 SS-1700, manufactured by Xenix Corporation, resolution 300 × 256 pixels) under the following conditions.

<e> The chicken meat was left as it was (thickness: 24 mm or 20 mm), and near-infrared light was irradiated onto the entire surface so that it penetrated in the thickness direction (no compression, no slits).
<f> Chicken meat (20 mm thick) was sandwiched between acrylic plates and compressed (compressed without crushing) to a thickness of 15 mm or 12 mm, and near-infrared light was irradiated over the entire surface so that it penetrated in the direction of compression (15 mm compression or 12 mm compression, no slits).
<g> Chicken meat (24 mm thick) was sandwiched between acrylic plates and compressed (compressed without crushing) to a thickness of 8 mm, and fixed in place. Near-infrared light was then locally irradiated through a 1 mm slit in the short direction formed in a metal plate so that it could be transmitted in the direction of compression (8 mm compression x 1 mm slit).
<h> Chicken meat (20 mm thick) was sandwiched between acrylic plates and compressed (compressed without crushing) to a thickness of 15 mm or 12 mm, and fixed in place. Near-infrared light was then locally irradiated through a 1 mm slit in the short direction formed in a metal plate so that it could be transmitted in the compression direction (15 mm compression or 12 mm compression x 1 mm slit).

The images obtained are shown in Figure 4. The top and bottom left images in Figure 4 are images for <e>, the two images in the lower center are images for <f>, the top right image is an image for <g>, and the two images in the lower right image are images for <h>. The results show that bone was difficult to detect in <e>, in which near-infrared light was irradiated over the entire surface without slits or compression and allowed to pass through, and in <f>, in which near-infrared light was irradiated over the entire surface with compression but without slits and allowed to pass through. However, bone (<g> is cartilage, <h> is bone; both are circled in the images) was clearly easier to detect in <g>, in which the specimen was compressed to 33.3% of its thickness, fixed, and then irradiated locally with near-infrared light with slits and allowed to pass through, and in <h>, in which the specimen was compressed to 75% or 60% of its thickness, fixed, and then irradiated locally with near-infrared light with slits and allowed to pass through.
Although images are not shown, even when near-infrared light was locally irradiated and transmitted through a 1 mm slit without compression, bones were easier to detect than in <e> and <f>.

Example 3
Prune pulp (dried product, thickness 18 mm, 17 mm, or 20 mm) containing seed fragments as foreign matter was irradiated with near-infrared light at a wavelength of 1050 nm using an LED light (UPD2450W-2S, manufactured by U-Technology), and the pulp that had been transmitted through this near-infrared light was photographed with an InGaAs camera (XEVA-165 SS-1700, manufactured by Xenix, resolution 300 × 256 pixels) under the following conditions, with the pulp, light source, slit, and camera all fixed and with a camera exposure time of 5000 μs. Note that compression under the following conditions was performed without crushing.

<i> Near-infrared light was irradiated so that it penetrated in the thickness direction (compression direction) under four conditions: 18 mm thick fruit pulp was left as is (no compression), compressed and fixed to a thickness of 14 mm (14 mm adjustment), compressed and fixed to a thickness of 12 mm (12 mm adjustment), and compressed and fixed to a thickness of 12 mm with a 10 mm wide slit formed in a paper plate (12 mm adjustment + slit).
<k> Near-infrared light was irradiated so that it penetrated in the thickness direction (compression direction) under four conditions: 17 mm thick fruit pulp was left as is (no compression), compressed and fixed to a thickness of 14 mm (14 mm adjustment), compressed and fixed to a thickness of 12 mm (12 mm adjustment), and compressed and fixed to a thickness of 12 mm with a 10 mm wide slit formed in a paper plate (12 mm adjustment + slit).
<l> Near-infrared light was irradiated so that it penetrated in the thickness direction (compression direction) under four conditions: 20 mm thick fruit pulp was left as is (no compression), compressed and fixed to a thickness of 14 mm (14 mm adjustment), compressed and fixed to a thickness of 12 mm (12 mm adjustment), and compressed and fixed to a thickness of 12 mm with a 10 mm wide slit formed in a paper plate (12 mm adjustment + slit).

The images obtained are shown in Figure 5. The top four images in Figure 5 are images <i>, the middle four images are images <k>, and the bottom four images are images <l>. These results show that for prune pulp of any thickness, compressing and fixing it to a thickness of 12 mm (60.0-70.6%), and then passing it through a slit and locally irradiating it with near-infrared light made the seed fragments clearer and easier to detect.

Example 4
A scallop adductor muscle (boiled, 10 mm thick) containing shell fragments and sea urchin spines as foreign matter was irradiated with near-infrared light at a wavelength of 1050 nm using an LED light (UPD2450W-2S, manufactured by U-Technology), and the adductor muscle that had been irradiated with near-infrared light was photographed using an InGaAs camera (XEVA-165 SS-1700, manufactured by Xenix, resolution 300 x 256 pixels) under the following conditions. Note that the compression under the following conditions was not crushed in any case.

<m> The adductor muscle was left in its original state (thickness 10 mm), and near-infrared light was irradiated over the entire muscle so that it penetrated in the thickness direction (no compression, no slit), or near-infrared light was irradiated locally through a slit with a width of 1 mm in the short direction formed in a metal plate so that it penetrated in the thickness direction (no compression x 1 mm slit).
<n> An adductor muscle (10 mm thick) is sandwiched between acrylic plates, compressed to a thickness of 9 mm, and fixed in place. Near-infrared light is then locally irradiated through a 5 mm or 1 mm slit in the short direction formed in the metal plate so that it can be transmitted in the direction of compression (compressed to 9 mm x 5 mm slit, compressed to 9 mm x 1 mm slit).
<p> An adductor muscle (10 mm thick) was sandwiched between acrylic plates and compressed to a thickness of 8 mm, and fixed in place. Near-infrared light was then locally irradiated through a 5 mm slit in the short direction formed in the metal plate so that it could be transmitted in the direction of compression (compressed to 8 mm x 5 mm slit).
<q> An adductor muscle (10 mm thick) was sandwiched between acrylic plates and compressed to a thickness of 7 mm, and fixed in place. Near-infrared light was then locally irradiated through a 5 mm slit in the short direction formed in the metal plate so that it could be transmitted in the direction of compression (compressed to 7 mm x 5 mm slit).
<r> An adductor muscle (10 mm thick) was sandwiched between acrylic plates and compressed to a thickness of 6 mm, and fixed in place. Near-infrared light was then locally irradiated through a 5 mm or 1 mm slit in the short direction formed in the metal plate so that it could be transmitted in the direction of compression (compressed to 6 mm x 5 mm slit, compressed to 6 mm x 1 mm slit).

The images obtained are shown in Figure 6. Starting from the left column of Figure 6, they are: image <m>, image <n>, image <p>, image <q>, and image <r>. As a result, in the top image of <m>, in which near-infrared light was irradiated and transmitted through the entire scallop without a slit or compression, foreign matter (shell fragments and sea urchin spines) could not be detected except for the areas exposed from the adductor muscle. However, by using a 1 mm slit to locally irradiate and transmit near-infrared light, the entire foreign matter became detectable (circled in the bottom image of <m>). Furthermore, by compressing and fixing the scallop until the thickness was 90% or less, and then locally irradiating and transmitting near-infrared light with a slit, the entire foreign matter was clearly easier to detect (circled in the images of <n>, <p>, <q>, and <r>).

Example 5
Potatoes containing fiber strings as foreign matter (heat-treated at 90°C for 15 minutes, 19 mm thick) were irradiated with near-infrared light at a wavelength of 1050 nm using an LED light (UPD2450W-2S, manufactured by U-Technology), and the potatoes that were allowed to transmit this near-infrared light were then photographed using an InGaAs camera (XEVA-165 SS-1700, manufactured by Xenix, resolution 300 x 256 pixels) under the following conditions. Note that the compression under the following conditions was performed without crushing.

<s> The potato was left as it was (thickness 19 mm), and near-infrared light was irradiated onto the entire surface so that it penetrated in the thickness direction (no slits or compression). Images were taken with a camera exposure time of 10,000 μs or 20,000 μs.
<t> The potato was left as is (thickness 19 mm) and locally irradiated with near-infrared light that had passed through a slit in the metal plate with a width of 5 mm in the short direction so that it could be transmitted in the thickness direction (slit present, no compression). Images were taken with a camera exposure time of 60,000 μs.
<u> A potato (19 mm thick) was sandwiched between acrylic plates and compressed to a thickness of 15 mm, 13 mm, or 12 mm. The potato was then locally irradiated with near-infrared light that had passed through a 5 mm slit in the short direction formed in the metal plate so that it could be transmitted in the direction of compression (with slit and compression). Images were taken with a camera exposure time of 50,000 μs.
<v> A potato (19 mm thick) was sandwiched between acrylic plates and compressed to a thickness of 13 mm or 12 mm. The potato was then locally irradiated with near-infrared light that had passed through a 1 mm slit in the short direction formed in the metal plate so that it could be transmitted in the direction of compression (with slit and compression). The image was captured with a camera exposure time of 120,000 μs.

The images obtained are shown in Figure 7. The two images on the far left of Figure 7 are <s> images, the image in the upper center of Figure 7 is <t> image, the three images in the lower center of Figure 7 are <u> images, and the two images on the far right of Figure 7 are <v> images. As a result, it was difficult to detect foreign matter in the <s> image, in which near-infrared light was irradiated onto the entire surface and transmitted through without slits or compression. However, by using slits or compressing and fixing the potato until the thickness was 63.2-78.9% and then irradiating and transmitting near-infrared light locally with slits, it was clearly easier to detect foreign matter (the strings mixed in slightly to the left of the center of the potato in each image) (<t>, <u>, <v> images).

Example 6
As defective samples, 10 samples were prepared for learning and 50 samples were prepared for testing, with 5 mm x 5 mm seed fragments (foreign matter) embedded in the center of the dried prune pulp. 10 samples were prepared for learning and 50 samples were prepared for testing, with 50 samples of non-defective prune pulp containing no seed fragments.

These were then irradiated with near-infrared light at a wavelength of 1050 nm using an LED light (UPD2450W-2S, manufactured by U-Technology). In the conventional method (control method), the near-infrared light was irradiated entirely and transmitted, and images were taken with an InGaAs camera (XEVA-165 SS-1700, manufactured by Xenix, resolution 300 x 256 pixels). In the improved method (method of the present invention), near-infrared light was passed through a slit (10 mm wide) and transmitted locally, and images were taken with an InGaAs camera (XEVA-165 SS-1700, manufactured by Xenix, resolution 300 x 256 pixels), with the flesh, light source, slit, and camera all fixed, as in Example 3. The conditions for these measurements are shown in Table 1 below.

Five inspectors visually inspected a total of 50 images, including 25 images of defective products and 25 images of non-defective products, for the presence or absence of foreign matter. These inspectors had previously reviewed the images of the training samples (images of defective products and images of non-defective products) and learned to distinguish between them. The results are shown in Table 2 below. These results demonstrate that the accuracy rate of the method of the present invention was 30% or more higher than the control method, making it easier to detect seed fragments mixed into prune pulp. Furthermore, it is expected that the accuracy rate will be further improved if a machine learning model trained using such images of defective products and non-defective products as training data is used, and it is believed that even smaller seed fragments may be detected.

100 Foreign object detection device 11 Solid food 13 Foreign object 20 Near-infrared light irradiating unit 21 Light source 23 Near-infrared light 23a Emitted light 23b Transmitted light 31 Near-infrared light adjusting unit 31a Slit 41 Imaging unit 51 Detecting unit 51a Storage device 51b Arithmetic processing device 61 Control unit 71 Compression fixing means 73 Conveying means D Width of slit in short direction L Length of slit in long direction

Claims (8)

A foreign object detection device that detects foreign objects mixed in solid food and having a near-infrared light transmittance lower than that of the solid food,
a compressing and fixing means for applying a compressive load to the solid food that is smaller than a crushing compressive load that crushes the solid food, thereby compressing and fixing the thickness of the solid food in a direction through which near-infrared light is transmitted;
a near-infrared light irradiator that locally irradiates the compressed solid food with near-infrared light;
an imaging unit that captures an image of the compressed solid food by transmitting the near-infrared light ;
a detection unit that detects the foreign matter from the captured image,
Foreign object detection device. The foreign object detection device of claim 1, wherein the near-infrared light irradiator comprises a light source that emits the near-infrared light, and a near-infrared light adjuster that adjusts the amount and/or direction of the near-infrared light that is emitted from the light source and passes through the solid food. The foreign object detection device of claim 2, wherein the near-infrared light adjustment unit is disposed between the light source and the solid food and includes a slit that narrows a portion of the light direction of the near-infrared light emitted from the light source. 4. The foreign object detecting device according to claim 1 , wherein the near-infrared light has a wavelength of 800 nm or more and 1200 nm or less. The foreign object detection device according to any one of claims 1 to 4, wherein the detection unit detects the foreign object from the images captured by the imaging unit using a machine learning model trained by providing data of defective images of the solid food containing the foreign object and good images of the solid food not containing the foreign object, or a machine learning model trained by providing data of the good product images. A foreign object detection device as described in any one of claims 1 to 5 , wherein the solid food is either livestock meat, seafood meat, fruit pulp, or potatoes, and the foreign object is one or more selected from the group consisting of bones, shells, seeds, and soft foreign objects. A method for detecting a foreign object mixed in a solid food and having a near-infrared light transmittance lower than that of the solid food, comprising:
a compressing and fixing step using a compressing and fixing means that applies a compressive load to the solid food that is smaller than a crushing compressive load that is a compressive load at which the solid food is crushed, thereby compressing and fixing the thickness of the solid food in a direction through which near-infrared light is transmitted;
a near-infrared light irradiation step of locally irradiating the compressed solid food with near-infrared light;
an imaging step of capturing an image of the compressed solid food by transmitting the near-infrared light ;
a foreign object detection step of detecting the foreign object from the captured image.
Foreign object detection methods. a compressing and fixing step using a compressing and fixing means that applies a compressive load to the solid food that is smaller than a crushing compression load that is a compressive load at which the solid food is crushed, thereby compressing and fixing the thickness of the solid food in a direction through which near-infrared light is transmitted;
a near-infrared light irradiation step of locally irradiating the compressed solid food with near-infrared light;
an imaging step of capturing an image of the compressed solid food by transmitting the near-infrared light ;
a foreign object detection step of detecting a foreign object having a lower near-infrared light transmittance than the solid food from the captured image;
a foreign matter removal step of removing the detected foreign matter from the solid food or removing the solid food containing the foreign matter.
Methods for producing solid foods.