JP7510145B2 - Meat bone discrimination device and other discrimination methods, and meat bone discrimination programs - Google Patents

Description

The present invention relates to a discrimination device, such as a meat bone discrimination device, which distinguishes between bones and cartilage when a block of meat is cut, a discrimination method, such as a meat bone discrimination method, and a discrimination program, such as a meat bone discrimination program.

When a block of meat is cut, known methods and devices for distinguishing between bone and meat on the cut surface of the meat to distinguish between bone and cartilage include technology that uses X-rays to obtain positional information, and technology that uses infrared rays and visible light in combination.

JP 2018-64501 A JP 2018-66649 A

The invention described in Patent Document 1, "Device and method for identifying bones in meat," is "a bone position detection device that detects bones in the cut surface of meat, and is equipped with a first image capture unit 12 that irradiates the cut surface with infrared light having a wavelength of 1100 nm to 1700 nm and obtains a first image by incorporating the reflected light of the infrared light, a second image capture unit 18 that irradiates the cut surface with visible light and incorporates the reflected light of the visible light to obtain a second image, and a subtraction unit that performs subtraction processing between the first image and the second image to obtain a difference image in which the bone area is identified and displayed." It is also described as "enabling automatic identification of bones in the cut surface of meat."

The invention described in Patent Document 2, "Apparatus and method for identifying bones in meat," is "a bone identification apparatus 10 for detecting bones in a cut surface of meat, comprising a first image capture unit 16 for irradiating the cut surface with visible light and obtaining a first image with the reflected light of the visible light, a second image capture unit 22 for irradiating the cut surface with infrared light having a wavelength of 1100 nm to 1700 nm and obtaining a second image with the reflected light of the infrared light, and a binarization process for the first image based on the difference in the luminance value of the reflected light in the lean, bone, and fat parts of the cut surface, and a method for detecting the area of the fat part. a first binarization processing unit that extracts regions of the cut surface of the lean meat, the bone, and the fat, a second binarization processing unit that binarizes the second image based on the difference in the luminance value of the reflected light in the lean meat, the bone, and the fat on the cut surface, and extracts the regions of the bone and the fat, and a bone extraction unit that extracts the region of the bone by subtracting the region of the fat extracted by the first binarization processing unit from the region of the fat and the bone extracted by the second binarization processing unit." and "Enables automatic identification of bones on the cut surface of meat."

"The three components of the cut surface of meat, the lean part, the bone part and the fat part, each have different moisture and fat contents. On the other hand, the wavelength of 1100 nm to 1700 nm (hereinafter referred to as "wavelength A")
) has a water absorption peak around 1450 nm. Due to these differences, when infrared light or visible light is irradiated onto a cut surface of meat, the brightness value of the reflected light differs for the lean part, the bone part, and the fat part.
"The luminance values of the reflected light of the lean and fatty parts of the first image and the second image are the same, and the luminance values of the reflected light of the bone parts are different. Therefore, when the subtraction unit performs subtraction processing between the first image and the second image, a difference image in which only the bone parts are distinguishably displayed can be obtained."

"This makes it possible to automatically identify bones that exist on or near the cut surface of meat.
Furthermore, by using the obtained positional information of the bones in the deboning device, it becomes possible to automatically debon the bones that are present on or near the cut surface of the meat.
Here, "the bone part existing near the cut surface of meat" means a bone part that is at a distance from the cut surface that allows the bone part to be identified in some embodiments" (Patent Document 1 [0007]).

The inventions described in Patent Documents 1 and 2 utilize infrared or visible light to distinguish between bone and cartilage when a block of meat is cut, making it possible to automatically identify bones present on or near the cut surface of the meat.
However, image processing such as characterizing brightness and the like becomes necessary in later processes, which necessitates adjustments using the human eye, as well as binarization, edge clarity processing, and the need for "a first imaging unit that irradiates infrared light and collects the reflected light of the infrared light to obtain a first image, a second imaging unit that irradiates visible light onto the cut surface and collects the reflected light of the visible light to obtain a second image, and a subtraction unit that performs subtraction processing between the first image and the second image to obtain a difference image that identifies and displays the bone area." This poses the problem of complicating the work.

On the other hand, the inventors discovered the following regarding the first embodiment.
1. When cutting a block of meat, it is effective to distinguish bone from cartilage by irradiating it with ultraviolet light (wavelengths 10 - 400 nm), violet visible light (wavelengths 380 - 450 nm), or blue visible light (wavelengths 450 - 495 nm).
2. Under ultraviolet light (wavelengths 10 - 400 nm), the bones appear white. The fascia around the bones appears purple and leopard-striped amoeba-like.
3. Under violet visible light (wavelengths 380 - 450 nm), the bones appear whitish yellowish. The fascia around the bones appears as a light purple leopard-striped amoeba.
4. Under blue visible light (wavelengths 450 - 495 nm), bone appears yellow. The fascia surrounding the bone has no detectable color or shape.

5. About filters Violet and blue visible light are clearly visible when viewed through an orange or yellow filter, making bones easier to distinguish from other objects.
6. The effect of ice The meat is frozen. Purple and blue visible light are not easily affected by ice. However, ultraviolet light is affected by ice and is diffused when irradiated. Therefore, by irradiating it in the dark with blue, purple, and infrared light in that order and overlaying the resulting images, it is effective in distinguishing between bone and cartilage.
7. Working environment It is best to work in a dark place. If it is not dark, the light emitted by the bones will be very weak, so it will be necessary to process the brightness on a computer, binarize the image, and perform edge processing (processing to clarify the edges).

On the other hand, the inventors further discovered the following.
Second Example: Yellow fishing line is put into mozuku seaweed. The fishing line emits strong fluorescence.
(1) Irradiate with a combination of visible light (wavelength 400 nm-700 nm) and ultraviolet light (wavelength 10-400 nm).
The fishing line appears to fluoresce, and the white conveyor belt underneath is visible through the gaps.
(2) Irradiate with ultraviolet light (wavelength 10-400 nm).
The fishing line is visible.
When exposed to only visible light, fishing line does not glow easily.

Third Example: A green fishing line is placed inside a piece of mozuku seaweed. The fluorescent light emitted by the fishing line is very weak.
(1) Irradiate with a combination of visible light (wavelength 400 nm-700 nm) and ultraviolet light (wavelength 10-400 nm).
Fishing lines are visible. Pieces of fishing net are visible.
(2) Irradiate with ultraviolet light (wavelength 10-400 nm).
There is no fishing line in sight.
When exposed to only visible light, fishing line does not glow easily.

Fourth Example: Brown fishing line is placed in mozuku seaweed. The brown fishing line emits fluorescence.
(1) Irradiate with a combination of visible light (wavelength 400 nm-700 nm) and ultraviolet light (wavelength 10-400 nm).
The fishing line is visible.
(2) Irradiate with ultraviolet light (wavelength 10-400 nm).
The fishing line is faintly visible.
When exposed to only visible light, fishing line does not glow easily.

Fifth Example: Shrimp is included in the mozuku seaweed. Fluorescence is emitted.
(1) Irradiate with a combination of visible light (wavelength 400 nm-700 nm) and ultraviolet light (wavelength 10-400 nm).
You can see the shrimp, and the white conveyor belt underneath can be seen through the gaps.
(2) Irradiate with ultraviolet light (wavelength 10-400 nm).
No shrimp or gaps to be seen.
Shrimp are not sensitive to light when exposed to only visible light.

Sixth Example: Fish are contained in mozuku seaweed. Fish emit fluorescence.
(1) Irradiate with a combination of visible light (wavelength 400 nm-700 nm) and ultraviolet light (wavelength 10-400 nm).
You can see the fish, and the white conveyor belt underneath, visible through the gaps.
(2) Irradiate with ultraviolet light (wavelength 10-400 nm).
The fish looks thin. The gap is invisible.
Fish do not glow easily when exposed to visible light alone.

Seventh Example: Shellfish are contained in mozuku seaweed. Shellfish emit fluorescence.
(1) Irradiate with a combination of visible light (wavelength 400 nm-700 nm) and ultraviolet light (wavelength 10-400 nm).
The shells are visible, as is the white conveyor belt underneath, visible through the gaps.
(2) Irradiate with ultraviolet light (wavelength 10-400 nm).
The shell looks thin. No gaps are visible.
Shells do not glow easily when exposed to visible light alone.

Eighth Example: Mozuku seaweed contains different kinds of seaweed. The different kinds of seaweed emit fluorescence.
(1) Irradiate with a combination of visible light (wavelength 400 nm-700 nm) and ultraviolet light (wavelength 10-400 nm).
Different species of seaweed are visible.
(2) Irradiate with ultraviolet light (wavelength 10-400 nm).
The different seaweeds appear thin.
When exposed to visible light only, different species of seaweed do not glow easily.

Ninth Example: Bean sprouts, a root vegetable, contain rotten roots. The rotten roots do not emit fluorescence.
(1) Irradiate visible light (wavelength 400 nm-700 nm).
The root rot areas are visible.
(2) Irradiate with visible light (red only) (wavelength 620-750 nm). Since there was no material that emitted fluorescence, red visible light was irradiated.
The rotted root areas appear thin.

Example 10: Bean sprouts, a root vegetable, contain damaged areas that have turned black. There is no fluorescence in the damaged areas.
(1) Irradiate visible light (wavelength 400 nm-700 nm).
The damaged area is visible. The stem is visible.
(2) Irradiate with visible light (red only) (wavelength 620-750 nm). Since there was no material that emitted fluorescence, red visible light was irradiated.
The blackened damaged areas appear lighter.

Eleventh Example: Bean sprouts, a root vegetable, contain damaged areas that have turned black. The damaged areas do not emit fluorescent light.
(1) Irradiate visible light (wavelength 400 nm-700 nm).
The damaged area is visible. The stem is visible.
(2) Irradiate with visible light (red only) (wavelength 620-750 nm). Since there was no material that emitted fluorescence, red visible light was irradiated.
The damaged areas that have turned black appear pale, and the stems appear dim.

Related to the 12th Example: Dried seaweed, called dried nori, contains a carbon fishing net. The fishing net does not emit fluorescent light.
(1) Irradiate near-infrared light (wavelength 0.7 - 2.5 μm).
Carbon fishing nets are visible.
Near-infrared light was applied from the side.
(2) Light other than near-infrared light was applied.

13th Example Carbon fishing nets are placed in dried wakame seaweed, which is a type of dried seaweed. Carbon fishing nets do not emit fluorescence.
(1) Irradiate near-infrared light (wavelength 0.7 - 2.5 μm).
The carbon fishing net is visible. Because near-infrared light was irradiated from the side, the side appears white and is easy to see.
Near-infrared light was applied from the side.
(2) Light other than near-infrared light was applied.

Example 14: There are tiny flying insects in the soba noodles. There is no fluorescence.
(1) Irradiate near-infrared light (wavelength 0.7 - 2.5 μm).
I can see tiny flying insects.
Near-infrared light was applied from the side. Because near-infrared light was applied from the side, the side appeared white and was easy to see.
(2) White light was applied.

15th Example: Transparent vinyl is included in the pork shreds. Transparent vinyl does not emit fluorescent light. However, it does cause diffuse reflection.
(1) Irradiate visible light (wavelength 400 nm-700 nm). Transparent vinyl causes diffuse reflection.
(2) Irradiate with visible light (red only) (wavelength 620-750 nm). Since there was no material that emitted fluorescence, red visible light was irradiated.

Example 16: Transparent vinyl is included in the pork shreds. Transparent vinyl does not emit fluorescent light. However, it does have diffuse reflection.
(1) Irradiate visible light (wavelength 400 nm-700 nm). There is diffuse reflection from a transparent vinyl bag.
(2) Irradiate with visible light (red only) (wavelength 620-750 nm). Since there was no material that emitted fluorescence, red visible light was irradiated.

17th Example: Transparent vinyl is included in the pork shreds. Transparent vinyl does not emit fluorescent light. However, it does cause diffuse reflection.
(1) Irradiate visible light (wavelength 400 nm-700 nm). Diffuse reflection occurs.
(2) Irradiate with visible light (red only) (wavelength 620-750 nm). Since there was no material that emitted fluorescence, red visible light was irradiated.

Related to the 18th Example Croquettes are fried foods containing peeled off batter or frozen parts of fried foods.
(1) Visible light (only blue light) (wavelength 450 - 495 nm) was applied.
(2) Visible light (red light only) (wavelength 620-750 nm) was irradiated.
(3) Visible light (white light) was applied.
(4) Visible light (only green light) (wavelength 495-570 nm) was irradiated.

Nineteenth Example A cream croquette is a fried food containing peeled off batter or frozen parts of fried food.
(1) Visible light (only blue light) (wavelength 450 - 495 nm) was applied.
(2) Visible light (red light only) (wavelength 620-750 nm) was irradiated.
(3) Visible light (white light) was applied.
(4) Visible light (only green light) (wavelength 495-570 nm) was irradiated.

Related to Example 20: Fried chicken, which is a fried food with a mixture of peeled off batter and frozen parts.
(1) Visible light (only blue light) (wavelength 450 - 495 nm) was applied.
(2) Visible light (red light only) (wavelength 620-750 nm) was irradiated.
(3) Visible light (only green light) (wavelength 495-570 nm) was irradiated.
(4) Visible light (blue plus white) was applied.

By applying different light to each area, it is possible to increase the accuracy of detection.

The present invention relates to
In the dark,
an ultraviolet light photographing unit that irradiates ultraviolet light onto a cut surface of meat and obtains an ultraviolet image by collecting reflected light of the ultraviolet light;
a purple light photographing unit that irradiates a cut surface of meat with purple light and obtains a purple light image by collecting reflected light of the purple light;
a blue light photographing unit that irradiates blue light onto a cut surface of meat and captures reflected light of the blue light to obtain a blue light image;
A meat bone discrimination device that discriminates a bone part of a cut surface of meat when extracting a bone , which is an extraction target, from image data obtained from the above,
A partitioning means for partitioning the image data;
an image classification means for calculating a probability that each of the segments of the image data divided by the segment division means is the object to be extracted based on the machine learning result;
a classification result conversion means for converting the probability that the image is the extraction target for each of the categories, which is calculated by the image classification means, into a pixel value;
an image processing means for performing image processing on an image having the pixel values converted by the classification result conversion means ,
a meat bone discriminating device which, when irradiated with ultraviolet light, judges a part which looks white to be a bone, when irradiated with purple visible light, judges a part which looks whitish-yellow to be a bone, and when irradiated with blue visible light, judges a part which looks yellow to be a bone;
It consists of:

The present invention further comprises:
The image processing means performs image processing on the image data divided by the division means,
The image identification means comprises a meat bone discrimination device, characterized in that the image data processed by the image processing means is subjected to calculation of a probability of the image data being a bone , which is the extraction target, for each of the categories of the image data.

The present invention further comprises:
The meat bone discrimination device is characterized in that an image is obtained by passing an orange or yellow filter through an ultraviolet light photographing section.

The present invention further comprises:
The meat bone discriminating device is characterized in that an image is obtained by passing an orange or yellow filter through a blue light photographing section.

The present invention relates to
In the dark,
an ultraviolet light photographing unit that irradiates ultraviolet light onto a cut surface of meat and obtains an ultraviolet image by collecting reflected light of the ultraviolet light;
a purple light photographing unit that irradiates a cut surface of meat with purple light and obtains a purple light image by collecting reflected light of the purple light;
a blue light photographing unit that irradiates blue light onto a cut surface of meat and captures reflected light of the blue light to obtain a blue light image;
obtained from
A meat bone discrimination method for discriminating a bone portion of a cut surface of meat when extracting a bone as an extraction target from image data, comprising the steps of:
A partitioning step of partitioning the image data;
an image classification step of calculating, for each of the segments of the image data divided by the processing of the segment division step, a probability that the segment is the object to be extracted, based on a machine learning result of the feature amount of the object to be extracted by a deep learning technique;
a classification result conversion step of converting the probability of the object being the extraction target for each of the categories, calculated by the processing of the image classification step, into a pixel value;
an image processing step of performing image processing on an image having pixel values converted by the processing of the classification result conversion step;
a determining step of determining that a portion that appears white is a bone when irradiated with ultraviolet light, that a portion that appears whitish-yellow is a bone when irradiated with purple visible light, and that a portion that appears yellow is a bone when irradiated with blue visible light;
The present invention relates to a method for identifying meat bones , the method comprising the steps of:

The present invention further comprises:
The method comprises obtaining an image through an orange or yellow filter on a purple light photographing section.

The present invention further comprises:
The method comprises obtaining an image of the meat or bone through an orange or yellow filter on a blue light photographing section.

The present invention relates to
In the dark,
an ultraviolet light photographing unit that irradiates ultraviolet light onto a cut surface of meat and obtains an ultraviolet image by collecting reflected light of the ultraviolet light;
a purple light photographing unit that irradiates a cut surface of meat with purple light and obtains a purple light image by collecting reflected light of the purple light;
a blue light photographing unit that irradiates blue light onto a cut surface of meat and captures reflected light of the blue light to obtain a blue light image;
obtained from
A program for causing a computer to execute a meat bone discrimination process for discriminating a bone portion of a cut surface of meat when extracting a bone as an extraction target from image data, comprising:
A partitioning step of partitioning the image data;
an image classification step of calculating, for each of the segments of the image data divided by the processing of the segment division step, a probability that the segment is the object to be extracted, based on a machine learning result of the feature amount of the object to be extracted by a deep learning technique;
a classification result conversion step of converting the probability of the object being the extraction target for each of the categories, calculated by the processing of the image classification step, into a pixel value;
an image processing step of performing image processing on an image having the pixel values converted by the processing in the classification result conversion step;
a determining step of determining that a portion that appears white is a bone when irradiated with ultraviolet light, that a portion that appears whitish-yellow is a bone when irradiated with purple visible light, and that a portion that appears yellow is a bone when irradiated with blue visible light;
The program comprises:

By irradiating a block of meat with ultraviolet light, purple visible light, and blue visible light, it is possible to identify bones present on or near the cut surface of the meat when the block is cut.
To distinguish between fishing line, shrimp, fish, shellfish, or different kinds of seaweed in seaweed in which fishing line, shrimp, fish, shellfish, or different kinds of seaweed are mixed.
To enable discrimination of root rot or damaged parts in root vegetables such as bean sprouts containing root rot or damaged parts.
To make it possible to distinguish carbon fishing nets from dried seaweed such as dried nori or dried wakame seaweed mixed with carbon fishing nets.
To enable discrimination of extremely small flying insects in buckwheat containing extremely small flying insects.
This makes it possible to distinguish transparent vinyl when it is mixed in with pork shreds.
To make it possible to distinguish peeled or frozen parts of fried foods in which peeled or frozen parts of the coating are mixed.

1 is a front view of a first embodiment of the present invention in use; FIG. 1 is a cross-sectional view showing bone and cartilage coexisting when a block of meat irradiated with visible light in a first embodiment of the present invention is cut. FIG. 1 is a cross-sectional view showing bone and cartilage coexisting when a block of meat irradiated with violet visible light in a first embodiment of the present invention is cut. FIG. 1 is a cross-sectional view showing bone and cartilage coexisting when a block of meat irradiated with violet visible light in a first embodiment of the present invention is cut. FIG. 1 is a cross-sectional view showing bone and cartilage coexisting when a block of meat irradiated with violet visible light in a first embodiment of the present invention is cut. FIG. 1 is a cross-sectional view showing bone and cartilage coexisting when a block of meat irradiated with violet visible light in a first embodiment of the present invention is cut. FIG. 1 is a cross-sectional view showing bone and cartilage coexisting when a block of meat irradiated with violet visible light in a first embodiment of the present invention is cut. FIG. 1 is a cross-sectional view showing bone and cartilage coexisting when a block of meat irradiated with violet visible light in a first embodiment of the present invention is cut. FIG. 1 is a cross-sectional view showing bone and cartilage coexisting when a block of meat irradiated with violet visible light in a first embodiment of the present invention is cut. FIG. FIG. 2 is a conceptual diagram of a cut surface of purple fascia of a meat block according to the first embodiment of the present invention. FIG. 2 is a functional block diagram for explaining functions of an image analyzing device according to a first embodiment of the present invention. 4 is a flowchart for explaining an image analysis process according to the first embodiment of the present invention. 4 is a flowchart for explaining an image identification process according to the first embodiment of the present invention; FIG. 11 is a diagram for explaining processing by a classifier having learning results regarding the cut surfaces of meat when a block of meat, which is an extraction target, is cut using a deep learning technique in the first embodiment which is an embodiment of the present invention. FIG. 11 is a diagram showing the appearance of meat parts at a cut surface where bone and cartilage coexist when a block of meat is cut after being irradiated with ultraviolet light, purple visible light, and blue visible light in the first embodiment of the present invention. 13 is a diagram showing the appearance of mozuku seaweed mixed with fishing line (yellow) when irradiated with ultraviolet light and visible light in a second embodiment of the present invention. FIG. FIG. 11 is a diagram showing the appearance of mozuku seaweed mixed with fishing line (yellow) irradiated with ultraviolet light in a second embodiment according to the present invention. 13 is a diagram showing the appearance of mozuku seaweed mixed with fishing line (green) when irradiated with ultraviolet light and visible light in a third embodiment of the present invention. FIG. FIG. 13 is a diagram showing the appearance of mozuku seaweed mixed with fishing line (green) irradiated with ultraviolet light in a third embodiment according to the present invention. FIG. 13 is a diagram showing the appearance of mozuku seaweed mixed with fishing line (brown) when irradiated with ultraviolet light and visible light in a fourth embodiment of the present invention. FIG. 13 is a diagram showing the appearance of mozuku seaweed mixed with fishing line (brown) irradiated with ultraviolet light in a fourth embodiment of the present invention; FIG. 13 is a diagram showing the appearance of mozuku seaweed containing shrimp when irradiated with ultraviolet light and visible light in a fifth embodiment of the present invention. FIG. 13 is a diagram showing the appearance of mozuku seaweed containing shrimp irradiated with ultraviolet light in a fifth embodiment according to the present invention. FIG. 13 is a diagram showing the appearance of mozuku seaweed, which is a type of seaweed mixed with fish, when irradiated with ultraviolet light and visible light in a sixth embodiment of the present invention; FIG. 13 is a diagram showing the appearance of mozuku seaweed, which is a type of seaweed mixed with fish, when irradiated with ultraviolet light in a sixth embodiment which is an embodiment of the present invention. 13 is a diagram showing the appearance of mozuku seaweed, a type of seaweed containing shellfish, when irradiated with ultraviolet light and visible light in the seventh embodiment of the present invention; FIG. 13 is a diagram showing the appearance of mozuku seaweed, a type of seaweed mixed with shellfish, irradiated with ultraviolet light in the seventh embodiment of the present invention. FIG. 13 is a diagram showing the appearance of mozuku seaweed, a type of seaweed containing a mixture of different seaweeds, when irradiated with ultraviolet light and visible light in the eighth embodiment of the present invention. This is a diagram showing the appearance of mozuku, a type of seaweed containing a mixture of different types of seaweed, when irradiated with ultraviolet light in the eighth embodiment which is an embodiment of the present invention. FIG. 13 is a diagram showing the appearance of bean sprouts, a root vegetable having root rot, irradiated with visible light in a ninth embodiment of the present invention. FIG. 13 is a diagram showing the appearance of bean sprouts, a root vegetable having root rot, when irradiated with visible light (only red light) in a ninth embodiment of the present invention. FIG. 19 is a diagram showing the appearance of bean sprouts, a root vegetable having damaged areas, when irradiated with visible light in the tenth embodiment of the present invention. This is a diagram showing the appearance of a bean sprout having a damaged area when irradiated with visible light (only red light) in the tenth embodiment of the present invention. FIG. 15 is a diagram showing the appearance of bean sprouts, a root vegetable having damaged areas, when irradiated with visible light in the eleventh embodiment of the present invention. FIG. 15 is a diagram showing the appearance of bean sprouts, a root vegetable having damaged areas, when irradiated with visible light (only red light) in the eleventh embodiment of the present invention. This figure shows the appearance of dried seaweed, which is a type of dried seaweed mixed with carbon fishing nets, when irradiated with near-infrared light in the 12th embodiment, which is an embodiment of the present invention. This is a figure showing the appearance of dried seaweed mixed with carbon fishing nets irradiated with near-infrared light in the 12th embodiment of the present invention. This figure shows the appearance of dried seaweed, which is a type of dried seaweed mixed with carbon fishing nets, when irradiated with near-infrared light in the 12th embodiment, which is an embodiment of the present invention. This figure shows the appearance of dried seaweed, which is a type of dried seaweed mixed with carbon fishing nets, when irradiated with near-infrared light in the 12th embodiment, which is an embodiment of the present invention. This figure shows the appearance of dried seaweed, which is a type of dried seaweed mixed with carbon fishing nets, when irradiated with near-infrared light in the 12th embodiment, which is an embodiment of the present invention. This figure shows the appearance of dried seaweed, which is a type of dried seaweed mixed with carbon fishing nets, when irradiated with near-infrared light in the 12th embodiment, which is an embodiment of the present invention. This figure shows the appearance of dried wakame seaweed, which is a type of dried seaweed mixed with carbon fishing nets irradiated with near-infrared light in the 13th embodiment, which is an embodiment of the present invention. This figure shows the appearance of dried wakame seaweed, which is a type of dried seaweed mixed with carbon fishing nets irradiated with near-infrared light in the 13th embodiment, which is an embodiment of the present invention. This figure shows the appearance of dried wakame seaweed, which is a type of dried seaweed mixed with carbon fishing nets irradiated with near-infrared light in the 13th embodiment, which is an embodiment of the present invention. This figure shows the appearance of dried wakame seaweed, which is a type of dried seaweed mixed with carbon fishing nets irradiated with near-infrared light in the 13th embodiment, which is an embodiment of the present invention. This is a diagram showing the appearance of buckwheat noodles containing tiny flying insects when irradiated with near-infrared light in the 14th embodiment of the present invention. This is a diagram showing the appearance of buckwheat noodles containing tiny flying insects when irradiated with near-infrared light in the 14th embodiment of the present invention. This is a diagram showing the appearance of buckwheat noodles containing tiny flying insects when irradiated with near-infrared light in the 14th embodiment of the present invention. This is a diagram showing the appearance of buckwheat noodles containing tiny flying insects when irradiated with near-infrared light in the 14th embodiment of the present invention. This is a diagram showing the appearance of pork pieces and transparent vinyl when irradiated with visible light in the 15th embodiment, which is an embodiment of the present invention. This is a diagram showing the appearance of pork pieces and transparent vinyl when illuminated with visible light (only red light) in the 15th embodiment of the present invention. This is a diagram showing the appearance of pork pieces and transparent vinyl when irradiated with visible light in the 16th embodiment, which is an embodiment of the present invention. This figure shows the appearance of pork pieces and transparent vinyl when illuminated with visible light (only red light) in the 16th embodiment, which is an embodiment of the present invention. This is a diagram showing the appearance of pork pieces and transparent vinyl when irradiated with visible light in the 17th embodiment of the present invention. This figure shows the appearance of pork pieces and transparent vinyl when illuminated with visible light (only red light) in the 17th embodiment of the present invention. This is a diagram showing a fried croquette irradiated with visible light (only blue light) in the 18th embodiment of the present invention, and the appearance of peeling coating or frozen areas of the fried food. This is a diagram showing a fried croquette irradiated with visible light (only red light) in the 18th embodiment of the present invention, and the appearance of peeling coating or frozen areas of the fried food. This is a diagram showing a fried croquette irradiated with visible light (only white light) in the 18th embodiment of the present invention, and the appearance of peeling coating or frozen areas of the fried food. This is a diagram showing a fried croquette irradiated with visible light (only green light) in the 18th embodiment of the present invention, and the appearance of peeling coating or frozen areas of the fried food. This is a diagram showing a fried cream croquette irradiated with visible light (only blue light) in the 19th embodiment of the present invention, and the appearance of peeling coating or frozen areas of the fried food. This is a diagram showing a fried cream croquette irradiated with visible light (only red light) in the 19th embodiment of the present invention, and the appearance of peeling coating or frozen areas of the fried food. This is a diagram showing a fried cream croquette irradiated with visible light (only white light) in the 19th embodiment of the present invention, and the appearance of peeling coating or frozen areas of the fried food. This is a diagram showing a fried cream croquette irradiated with visible light (only green light) in the 19th embodiment of the present invention, and the appearance of peeling coating or frozen areas of the fried food. This figure shows fried chicken irradiated with visible light (only blue light) in the twentieth embodiment of the present invention, and the appearance of peeling coating or frozen areas of the fried chicken. This figure shows fried chicken irradiated with visible light (only red light) in the twentieth embodiment of the present invention, and the appearance of peeling coating or frozen areas of the fried chicken. This figure shows fried chicken irradiated with visible light (only green light) in the 20th embodiment of the present invention, and the appearance of peeling coating or frozen areas of the fried chicken. This figure shows fried chicken irradiated with visible light (only blue light) in the twentieth embodiment of the present invention, and the appearance of peeling coating or frozen areas of the fried chicken.

An image analyzing apparatus according to a first embodiment of the present invention will be described below with reference to FIGS.
FIG. 11 is a functional block diagram for explaining the functions of the image analyzing device 1. As shown in FIG.

The image analysis device 1 has an irradiation unit 11, an image data acquisition unit 21, an operation input information acquisition unit 12, an image data division processing unit 13, an image recognition processing unit 14, a recognition result conversion processing unit 15, an image processing unit 16, an output processing unit 17, and a memory unit 18.
The irradiation unit 11 is composed of an ultraviolet light irradiation unit 111 constituting an ultraviolet light photographing unit, a purple light irradiation unit 112 constituting a purple light photographing unit, and a blue light irradiation unit 113 constituting a blue light photographing unit. In this embodiment, the purple light irradiation unit 112 uses purple visible light (910 nm). In this embodiment, the blue light irradiation unit 113 uses blue visible light (950 nm).
The image data acquisition unit 21 includes an ultraviolet light image data acquisition unit 211, a violet light image data acquisition unit 212, and a blue light image data acquisition unit 213, which constitute an ultraviolet light imaging unit.
The image data acquisition unit 21, which constitutes an ultraviolet light photographing unit, a violet light photographing unit, and a blue light photographing unit, receives reflected light from the irradiation unit 11. Reference numeral 221 denotes a filter. The filter 221 is an orange or yellow filter.

The image data acquisition unit 21 acquires image data of an object to be inspected, captured by, for example, a digital still camera, a line sensor camera, or the like, and supplies the image data to the image data division processing unit 13 .
An image is obtained by superimposing the image received with violet light, the image received with blue light, and the image received with infrared light shown in FIGS.

The present invention provides an apparatus and method for detecting bones by using ultraviolet light, violet visible light, and blue visible light in combination, and for distinguishing between bones and cartilage when a block of meat is cut. The present invention provides an apparatus and method for distinguishing between bones and cartilage when a block of meat is cut, and an embodiment of a meat bone discrimination apparatus, a meat bone discrimination method, and a meat bone discrimination program.
3 to 9 are cross-sectional views of a block of meat irradiated with violet visible light according to an embodiment of the present invention, in which bone and cartilage coexist when the block is cut. A, B, C, and D are bones, E is cartilage, F is a bundle of muscle fibers, G is endomysium, and H is perimysium.
In Fig. 10, which is a conceptual diagram of a cut surface of the purple fascia according to an embodiment of the present invention, F is a bundle of muscle fibers, G is an endomysium, H is a perimysium, and I is an epimysium.
FIG. 2 shows a cut surface of a block of meat W, in which bones A, B, C, D and cartilage E coexist, after natural light is applied and received without filtering.
Bone C, bone D, and cartilage E appear white.

(1) How meat appears when using LED UV light, violet visible light, and blue visible light

Ultraviolet (wavelength 10 - 400 nm)
As shown in Figure 15, no transmission filter is used for receiving ultraviolet light. Under ultraviolet light, ice appears as white bands. Red meat appears dark red, but has no detectable shape. Fat appears white, but has no detectable shape. Bones A, B, C, and D appear as floating islands of white. Cartilage E appears as floating islands of white. Muscles appear as white stripes. Fascia appears purple and ameba-like. A cross section of fascia resembles a leopard print.
Cartilage E is located near the side of the fascia. By knowing the presence of the fascia, the location of the cartilage E can be determined.

Purple (wavelength 380 - 450 nm)
As shown in Fig. 15, in Figs. 3 to 9 which are cross-sectional views of bones A, B, C, and D and cartilage E when a block of meat irradiated with violet visible light according to an embodiment of the present invention is cut, fat appears white, but has no detectable shape. Bones A, B, C, and D are whitish yellow and appear like floating islands. Cartilage E is pale white and appears like floating islands. Muscles appear as whitish yellow stripes. Fascia appears as purple and amoeba-like.
In this embodiment, violet visible light (910 nm) is used, but a yellow or orange filter is used to receive violet visible light. With violet visible light, ice has no detectable characteristics in terms of color or shape. Red meat appears reddish-black, but has no detectable shape.

Blue (wavelength 450 - 495 nm)
As shown in FIG. 15, in this embodiment, blue visible light (950 nm) is used, but a yellow or orange filter is used to receive blue visible light.
Under blue visible light, ice has no detectable characteristics in either color or shape. Red meat appears dark red but has no detectable characteristics in shape. Fat appears white but has no detectable characteristics in shape. Bones A, B, C, and D are yellow and appear like floating islands. Cartilage E is pale white and appears like floating islands. Muscles appear yellow and streaky but have no detectable characteristics in either color or shape.

(2) How to detect bones: Ultraviolet, violet, and blue light are all detected as small dots like floating islands, so the AI can learn their color and shape (see photo).

(3) How to detect cartilage: Ultraviolet, violet, and blue light all detect cartilage as small dots like floating islands. However, bones A, B, C, and D can be distinguished because fascia, which is detected using ultraviolet and violet light, is nearby. The AI can then learn that there is a high probability of cartilage being present near fascia.

(4) How to detect streaks: Ultraviolet, purple, and blue light are all detected as streaks, so the AI can learn their colors and shapes.
Ultraviolet, violet and blue light images are superimposed.
Since bones A, B, C, D and cartilage E are the same color, they are identified by shape using AI.

(5) How to detect fascia: When exposed to ultraviolet and purple light, it is detected as a purple, leopard-stained amoeba-like shape, so the AI can learn the color and shape.
The fascia is exposed in the cross section when bones are removed from muscles. Therefore, it is closely related to the position of the bones, and by having AI learn the position of the fascia, it can be used as a reference for detecting the positions of bones A, B, C, D, and cartilage E. Looking at the muscle structure in the figure below, it is easy to see why it can be detected as a leopard print.

(6) Other points to note for detection When using the three types of LEDs to detect bones, it is necessary to work in a dark place. It is impossible in bright surroundings. If you use a neutral white LED in combination with this or try to detect under fluorescent lighting, the luminous color of the bones will be very weak, and the luminescence that identifies the bones will vary too much. This requires image processing to characterize the brightness in the post-process, which requires adjustments using the human eye. Binarization is not necessary.
It is difficult to detect bones using color and shape without using a deep learning AI training method, so it is essential to use deep learning AI.
Deep learning, a branch of artificial intelligence (AI), enables computers to learn new concepts from raw data, just like humans can. A deep learning system can identify differences in a particular image after seeing thousands of different versions of it.

The effect of ice: The meat is frozen. Violet and blue light are not affected by ice. Ultraviolet light is affected by ice and is diffusely reflected.
In the dark, blue light, violet light, and ultraviolet light are shone on the object in that order. When the images of the blue light and violet light are viewed through an orange, yellow, or orange filter, they are clearly visible.
The obtained images are overlaid and judged using AI.

When meat is irradiated with infrared light, especially far infrared light (wavelengths of approximately 4-1000 μm), the meat parts become easier to see.
Near infrared rays have wavelengths of approximately 0.7 - 2.5 μm.
Mid-infrared has wavelengths of approximately 2.5 - 4 μm
Far infrared rays have a wavelength of approximately 4 - 1000 μm.
It is said.
In the present invention, infrared rays may also be used in combination.
An image received with violet light, an image received with blue light, and an image received with infrared light as shown in FIGS. 3 to 9 are obtained, and a superimposed image is obtained.

The operation input information acquisition unit 12 acquires the user's operation input input from the input unit 2, and supplies information corresponding to the user's operation input to each of the image data division processing unit 13, the image recognition processing unit 14, the recognition result conversion processing unit 15, the image processing unit 16, and the output processing unit 17.

The image data division processing unit 13 divides and segments the image data for an image obtained by superimposing an image received under purple light, an image received under blue light, and an image received under infrared light as shown in Figures 3 to 9, which is supplied from the image data acquisition unit 21, and supplies the divided images to the image identification processing unit 14 or the image processing unit 16 as square cells of a predetermined size.
The size of the cells divided by the image data division processing unit 13 can be specified by the user. The image data division processing unit 13 corresponds to a partition dividing means. Furthermore, the cells correspond to partitions divided by the partition dividing means. Furthermore, the cells are image areas of a predetermined size corresponding to the partitions divided by the image data division processing unit 13, and may be constituted by square areas of pixels having the same number of pixels vertically and horizontally, or may be constituted by rectangular areas of different vertical and horizontal sizes.

The image recognition processing unit 14 has the learning results of the extraction object using a deep learning method, and discovers the feature quantities of the extraction object from the supplied image data. For example, from many images in which bones A, B, C, D and cartilage E coexist, which are generated when a block of meat is cut, the image recognition processing unit 14 has learned the feature quantities to distinguish between the coexisting bones A, B, C, D and cartilage E, which are generated when the block of meat is cut, using a deep learning method, and is then able to identify the bones, which are the extraction object, from the supplied image data.

Deep learning is an information processing technology that mimics the network of nerve cells in the human brain (neural network) and is capable of discovering features contained in images, sounds, etc. based on the learning results by deepening the layers of a neural network (multi-layer network).
In conventional image and voice recognition and identification technologies, the user sets features, and then a classification algorithm is applied based on the features to classify the extraction target, but by using deep learning to have the image identification processing unit 14 perform machine learning, it is possible to extract the extraction target without the need for the user to set features. The image identification processing unit 14 calculates the probability that each cell constituting the image data supplied from the image data division processing unit 13 or the image processing unit 16 is the extraction target, and supplies it to the identification result conversion processing unit 15. The image identification processing unit 14 corresponds to an image identification means.

The identification result conversion processing unit 15 converts the probability that each cell constituting the image data supplied from the image identification processing unit 14 is an object to be extracted into a pixel value using a method described below, generates image data for each cell according to the converted pixel value, and supplies it to the image processing unit 16. The identification result conversion processing unit 15, for example, refers to a tone curve (or a lookup table) and converts the probability (0% to 100%) that each cell is an object into a pixel value having a value from 0 to 255. For example, if the probability that a cell is an object to be extracted is 100%, the pixel value is converted to 255 based on the tone curve. The identification result conversion processing unit 15 corresponds to an identification result conversion means.

The image processing unit 16 performs image processing on the image data supplied from the identification result conversion processing unit 15, which indicates the probability that the pixel value of each cell is the object to be extracted, or on the divided cell image data supplied from the image data division processing unit 13, such as noise removal using a threshold value, the processor's knowledge of distinguishing between coexisting bones A, B, C, D and cartilage E that arise when cutting a block of meat, expansion processing, closing processing, skeletal line extraction processing, and the like, and supplies the processed image data to the image identification processing unit 14 or the output processing unit 17. The image processing unit 16 corresponds to image processing means.

Here, the image data divided into cells by the image data division processing unit 13 may be processed first by either the image identification processing unit 14 or the image processing unit 16. Also, the image data divided into cells by the image data division processing unit 13 may be processed multiple times by the image identification processing unit 14 or the image processing unit 16. In the image analysis device 1, it is possible to set the order and number of times of processing in the image identification processing unit 14 and the image processing unit 16 so as to suit the input image data and the characteristics of the object to be extracted, based on the user's operation input input by the input unit 2.

The output processing unit 17 supplies the processed image data supplied from the image processing unit 16 to the output unit 3 for output, or to the storage unit 18 for storage. The output processing unit 17 also reads out the processed image data stored in the storage unit 18, and supplies it to the output unit 3 for output.

The storage unit 18 stores the processed image data supplied from the output processing unit 17 in a built-in storage medium or an attached removable storage medium.

In the image analysis device 1, the probability that each cell constituting the image data identified in the image identification processing unit 14 is an object to be extracted is converted into image data in the identification result conversion processing unit 15, and the image data is processed to detect the object to be extracted. Therefore, even if the overall brightness, etc. of images taken at different times differs, the analysis results are not affected by this, and it is suitable for use when comparing changes over time in the same part.

Next, referring to the flowchart in FIG. 12, as an example of processing executed by the image analysis device 1, image analysis processing 1 executed by the image analysis device 1 equipped with an image recognition processing unit 14 that has learned the feature amounts of a block of meat in which bones A, B, C, D and cartilage E coexist, which arise when a block of meat is cut using a deep learning technique, from many images of a block of meat in which bones A, B, C, D and cartilage E coexist, which arise when a block of meat is cut.

In step S1, the image data acquisition unit 21 acquires image data to be analyzed, for example, as shown in FIG. 11, and supplies the image data to the image data division processing unit 13.

In step S2, the image data division processing unit 13 divides the image data to be analyzed and supplies it to the image recognition processing unit 14. The size of one divided cell may be set by the user depending on the type of image data to be analyzed and the object to be extracted. One cell may be configured as a square of image data to be analyzed, for example, 64 pixels vertical by 64 pixels horizontal, or as a rectangle of different vertical and horizontal sizes. In addition, one cell may be, for example, a 1 cm square or a 5 cm square of a block of meat in which bones A, B, C, and D and cartilage E coexist, as occurs when an actual block of meat is cut. The image resolution per unit pixel can be freely set.

In step S3, the image recognition process described below with reference to FIG. 13 is executed.

In step S4, the classification result conversion processing unit 15 converts the classification result from the image classification process, i.e., the probability that each cell is an object to be extracted, calculated by the image classification processing unit 14, into a pixel value based on, for example, a tone curve, and supplies it to the image processing unit 16.

In step S5, conventional image processing 1, which will be described later with reference to FIG. 14, is executed.

In step S6, the output processing unit 17 supplies the processed image analysis results supplied from the image processing unit 16 to the storage unit 18. The storage unit 18 stores the supplied image analysis results. The output processing unit 17 also supplies the processed image analysis results supplied from the image processing unit 16 to the output unit 3 for output, and the processing is then terminated.

Through this processing, the probability that each cell that makes up the image data is the object to be extracted is converted into image data, and the image data is used to detect the object to be extracted. Therefore, even if the overall brightness of the image, etc., is different, the analysis results will not be affected.
Furthermore, the extraction results of the object obtained in this manner do not include parts that may be confused with bones, such as those that arise when cutting a block of meat, in which bones A, B, C, and D coexist with cartilage E, and are extracted together with the bones, making it possible to obtain analysis results that are close to those obtained by visual inspection.

Next, the image identification process executed in step S3 of FIG. 12 will be described with reference to the flowchart of FIG. 13.

In step S11, the image identification processing unit 14 receives a designation of a target to be identified based on a user's operation input supplied from the operation input information acquisition unit 12. Here, the image identification processing unit 14 receives a designation of a bone as the target to be identified.

In step S12, the image recognition processing unit 14 performs target recognition processing based on the learning results.

In step S13, the image recognition processing unit 14 calculates the probability that the pixel value of the pixel included in the target is the extraction target for each cell based on the recognition result by the processing in step S12, and supplies the calculated probability to the recognition result conversion processing unit 15, and the processing returns to step S4 in FIG. 12.

Through this type of processing, the image recognition processing unit 14 can use a deep learning technique to extract the object to be extracted while eliminating parts that may be confused with bones, without the need for the user to set features, based on the learning results of features obtained from many images of a block of meat in which bones A, B, C, and D and cartilage E coexist, for example, when a block of meat is cut.

Next, conventional image processing 1, which is executed in step S5 of FIG. 12, will be described with reference to the flowchart of FIG. 14.

In step S21, the image processing unit 16 sets the pixel value of the cell image having a pixel value equal to or less than a predetermined threshold to 0 for the cell image data having pixel values supplied from the identification result conversion processing unit 15, i.e., image data indicating the probability that the pixel value of the cell image is the object to be extracted. The image data supplied from the identification result conversion processing unit 15 is image data in which, when a block of meat is cut, a block of meat in which bones A, B, C, D and cartilage E coexist is generated, and cells in which the probability of bones being present is high are indicated by dots with high pixel values, and cells in which the probability of bones being present is low are indicated by dots with low pixel values, when a block of meat is cut, and cells in which bones A, B, C, D and cartilage E coexist is generated, and

When 70% of the pixel value is set as the threshold and cells with pixel values below this are set to 0, image data is obtained in which noise components due to parts that have characteristics similar to those of bones present in a block of meat in which bones A, B, C, and D coexist with cartilage E, which arise when cutting a block of meat, have been removed to a certain extent.

In step S22, the image processing unit 16 executes an expansion process to expand the area of the dots that have pixel values due to the process in step S21. This process shows the parts of the images in Figures 3 to 9, etc., where the bones present in the block of meat that contains bones A, B, C, and D and cartilage E are detected as being broken when the block of meat is cut.

In step S23, the image processing unit 16 detects the length of consecutive pixels from the image data obtained by the processing in step S22, and removes components whose length is equal to or less than a predetermined threshold. This processing makes it possible to obtain image data from the image data in Figures 3 to 9, etc., from which parts that could be mistaken for bones and noise components that could not be removed by the processing in step S21 have been removed.

In step S24, the image processing unit 16 performs a closing process to smooth the edge portions of the image data obtained by the process in step S23. This process makes it possible to obtain image data from the image data shown in Figures 3 to 9, etc.

In step S25, the image processing unit 16 performs a skeletal line extraction process on the image data obtained by the process of step S24, thereby returning the image data to a shape close to that of the bones present in the block of meat in which bones A, B, C, and D and cartilage E coexist, which is generated when the block of meat is cut and expanded by the expansion process of step S22, and thereby obtaining image data.

In step S26, the image processing unit 16 detects the length of consecutive pixels in the image data obtained by the processing of step S25, and removes components whose length is equal to or less than a predetermined threshold. This process removes from the image data any parts that could be mistaken for bones, and noise components that could not be removed by the processing of steps S21 and S23, and image data such as that shown in FIG. 13 can be obtained. After the processing of step S26 is completed, the process returns to step S6 in FIG. 12.

By performing this type of processing, the image processing unit 16 can extract the object to be detected more accurately than the image recognition results obtained by a classifier that has learning results of the object to be extracted using deep learning techniques, by further performing image processing, which has been conventionally performed, on the image data supplied from the recognition result conversion processing unit 15 and in which the pixel value of each dot indicates the probability that it is the object to be extracted.

In contrast, if the probability that each cell is an extraction target, detected using an image discrimination processing unit 14 with capabilities equivalent to a discriminator capable of outputting extraction results, is converted into image data by a discrimination result conversion processing unit 15, and processed by an image processing unit 16 with an image processing function that reflects knowledge about bones that exist in a block of meat in which bones A, B, C, D and cartilage E coexist, which arise when a block of meat is cut, it becomes possible to obtain extraction results in which noise components similar to bones that exist in a block of meat in which bones A, B, C, D and cartilage E coexist, which arise when a block of meat is cut, have been removed. In this way, it becomes possible to obtain analysis results that are close to those obtained by visual confirmation using the image analysis device 1.

The above-mentioned series of processes can be executed by hardware or by software. When executing the series of processes by software, the programs that make up the software are installed from a program recording medium into a computer that is built into dedicated hardware, or into a general-purpose personal computer, for example, that can execute various functions by installing various programs.

A second embodiment will be described below with reference to Figures 16 and 17. In the second embodiment, a yellow fishing line 16-2 is inserted into a piece of mozuku seaweed 16-1. The fishing line 16-2 emits strong fluorescence.
(1) In the state shown in FIG. 16 of the second embodiment, visible light (wavelength 400 nm to 700 nm) and ultraviolet light (wavelength 10 to 400 nm) are irradiated in combination.
The fishing line 16-2 appears to be fluorescent, and 16-3 is the white conveyor belt that is the background visible through the gap.
(2) In the state shown in FIG. 17 of the second embodiment, ultraviolet light (wavelength 10-400 nm) is irradiated.
The fishing line 16-2 is visible, but the white conveyor belt 16-3 is not visible.
When illuminated with only visible light, the fishing line 16-2 does not shine easily.

In the second embodiment, an image is obtained by superimposing images received using visible light and ultraviolet light, as well as multiple images received using ultraviolet light, as shown in Figures 16 and 17. These images are superimposed and identified by shape using AI.
The image analysis process executed by the image analysis device 1 using the image discrimination processing unit 14, which has learned the feature amount of a state in which a yellow fishing line 16-2 is inserted into a seaweed 16-1 and a gap 16-3 is captured using a deep learning method from multiple images, is performed in the same manner as in the first embodiment. Instead of the "meat/bone discrimination" in the first embodiment, discrimination between the seaweed 16-1 and the yellow fishing line 16-2 is performed according to [0065] to [0113]. Instead of the ultraviolet light irradiation unit 111 constituting the ultraviolet light imaging unit, the purple light irradiation unit 112 constituting the purple light imaging unit, and the blue light irradiation unit 113 constituting the blue light imaging unit constituting the first embodiment, a visible light imaging unit and an ultraviolet light imaging unit are provided.

A third embodiment will be described below, as shown in Figures 18 and 19. In the third embodiment, a green fishing line 18-2 is inserted into mozuku seaweed 18-1. The fluorescent light emitted by the fishing line 18-2 is weak.
(1) In the state shown in FIG. 18 of the third embodiment, visible light (wavelength 400 nm to 700 nm) and ultraviolet light (wavelength 10 to 400 nm) are irradiated in combination.
Fishing line 18-2 is visible. 18-3 is a piece of fishing net.
(2) In the state shown in FIG. 19 of the third embodiment, ultraviolet light (wavelength 10-400 nm) is irradiated.
Fishing line 18-2 is not visible.
When illuminated with only visible light, the fishing line 18-2 does not shine easily.

In the third embodiment, an image is obtained by superimposing images received using visible light and ultraviolet light, as well as multiple images received using ultraviolet light, as shown in Figures 18 and 19. These images are superimposed and identified by shape using AI.
The image analysis process executed by the image analysis device 1 using the image discrimination processing unit 14 that has learned the feature amount of the state in which the green fishing line 18-2 is inserted into the mozuku seaweed 18-1 using a deep learning method from multiple images is performed in the same manner as in the first embodiment. Instead of the "meat/bone discrimination" in the first embodiment, the discrimination between the mozuku seaweed 18-1 and the green fishing line 18-2 is performed according to [0065] to [0113]. Instead of the ultraviolet light irradiation unit 111 constituting the ultraviolet light imaging unit, the purple light irradiation unit 112 constituting the purple light imaging unit, and the blue light irradiation unit 113 constituting the blue light imaging unit constituting the first embodiment, a visible light imaging unit and an ultraviolet light imaging unit are provided.

A fourth embodiment will be described below with reference to Figures 20 and 21. In the fourth embodiment, a brown fishing line 20-2 is inserted in a mozuku seaweed 20-1. The brown fishing line 20-2 emits fluorescence.
(1) In the state shown in FIG. 20 of the fourth embodiment, visible light (wavelength 400 nm to 700 nm) and ultraviolet light (wavelength 10 to 400 nm) are irradiated in combination.
Fishing line 20-2 is visible.
(2) In the state shown in FIG. 21 of the fourth embodiment, ultraviolet light (wavelength 10-400 nm) is irradiated.
Fishing line 20-2 is faintly visible.
When only visible light is applied, the fishing line 20-2 does not easily glow.

In the fourth embodiment, an image is obtained by superimposing images received using visible light and ultraviolet light, as well as multiple images received using ultraviolet light, as shown in Figures 20 and 21. These images are superimposed and identified by shape using AI.
The image analysis process executed by the image analysis device 1 using the image discrimination processing unit 14 that has learned the feature amount of the state in which the brown fishing line 20-2 is inserted into the mozuku seaweed 20-1 using a deep learning method from multiple images is performed in the same manner as in the first embodiment. Instead of the "meat/bone discrimination" in the first embodiment, the mozuku seaweed 20-1 and the brown fishing line 20-2 are discriminated according to [0065] to [0113]. Instead of the ultraviolet light irradiation unit 111 constituting the ultraviolet light imaging unit, the purple light irradiation unit 112 constituting the purple light imaging unit, and the blue light irradiation unit 113 constituting the blue light imaging unit constituting the first embodiment, a visible light imaging unit and an ultraviolet light imaging unit are provided.

A fifth embodiment will be described below with reference to Figures 22 and 23. In the fifth embodiment, shrimp 22-2 are contained in mozuku seaweed 22-1. The mozuku seaweed emits fluorescence.
(1) In the state shown in FIG. 22 of the fifth embodiment, visible light (wavelength 400 nm to 700 nm) and ultraviolet light (wavelength 10 to 400 nm) are irradiated in combination.
Shrimp 22-2 is visible. 22-3 is the white conveyor belt that is the base visible through the gap.
(2) In the state shown in FIG. 23 of the fifth embodiment, ultraviolet light (wavelength 10-400 nm) is irradiated.
Shrimp 22-2 and gap 22-3 are not visible.
When irradiated with only visible light, the shrimp 22-2 does not easily glow.

In the fifth embodiment, an image is obtained by superimposing images received using visible light and ultraviolet light, and multiple images received using ultraviolet light, as shown in Figures 22 and 23. These images are superimposed and identified by shape using AI.
The image analysis process executed by the image analysis device 1 using the image discrimination processing unit 14 that has learned feature amounts of a state in which a shrimp 22-2 is inserted into a seaweed mozuku seaweed 22-1 and a gap 22-3 is captured using a deep learning method from a plurality of images is performed in the same manner as in the first embodiment. Instead of the "meat/bone discrimination" in the first embodiment, discrimination between the seaweed mozuku seaweed 22-1 and the shrimp 22-2 is performed according to [0065] to [0113]. Instead of the ultraviolet light irradiation unit 111 constituting the ultraviolet light imaging unit, the purple light irradiation unit 112 constituting the purple light imaging unit, and the blue light irradiation unit 113 constituting the blue light imaging unit constituting the first embodiment, a visible light imaging unit and an ultraviolet light imaging unit are provided.

A sixth embodiment will be described with reference to Figures 24 and 25. In the sixth embodiment, a fish 24-2 is contained in mozuku seaweed 24-1. The fish 24-2 emits fluorescence.
(1) In the state shown in FIG. 24 of the sixth embodiment, visible light (wavelength 400 nm to 700 nm) and ultraviolet light (wavelength 10 to 400 nm) are irradiated in combination.
The fish 24-2 is visible. 24-3 is the white conveyor belt background visible through the gap.
(2) In the state shown in FIG. 25 of the sixth embodiment, ultraviolet light (wavelength 10-400 nm) is irradiated.
The fish 24-2 is faintly visible, and the gap 24-3 is not visible.
When only visible light is irradiated, the fish 24-2 does not glow easily.

In the sixth embodiment, an image is obtained by superimposing images received using visible light and ultraviolet light, as well as multiple images received using ultraviolet light, as shown in Figures 24 and 25. These images are superimposed and identified by shape using AI.
The image analysis process executed by the image analysis device 1 using the image discrimination processing unit 14 that has learned feature amounts of a state in which a fish 24-2 enters a seaweed mozuku seaweed 24-1 and a gap 24-3 is captured using a deep learning method from multiple images is performed in the same manner as in the first embodiment. Instead of the "meat from bone discrimination" in the first embodiment, discrimination between the seaweed mozuku seaweed 24-1 and the fish 24-2 is performed according to [0065] to [0113]. Instead of the ultraviolet light irradiation unit 111 constituting the ultraviolet light imaging unit, the purple light irradiation unit 112 constituting the purple light imaging unit, and the blue light irradiation unit 113 constituting the blue light imaging unit constituting the first embodiment, a visible light imaging unit and an ultraviolet light imaging unit are provided.

A seventh embodiment will be described below with reference to Figures 26 and 27. In the seventh embodiment, a shell 26-2 is contained in a mozuku seaweed 26-1. The shell 26-2 emits fluorescence.
(1) In the state shown in FIG. 26 of the seventh embodiment, visible light (wavelength 400 nm to 700 nm) and ultraviolet light (wavelength 10 to 400 nm) are irradiated in combination.
Shell 26-2 is visible. 26-3 and 26-4 are the white conveyor belt background visible through the gaps.
(2) In the state shown in FIG. 27 of the seventh embodiment, ultraviolet light (wavelength 10-400 nm) is irradiated.
The shell 26-2 is thin and the gaps 26-3 and 26-4 are not visible.
When only visible light is irradiated, the shell 26-2 does not shine easily.

In the seventh embodiment, images are obtained by superimposing images received using visible light and ultraviolet light, as well as multiple images received using ultraviolet light, as shown in Figures 26 and 27. These images are superimposed and identified by shape using AI.
The image analysis process executed by the image analysis device 1 using the image discrimination processing unit 14 that has learned feature amounts of a state in which a shellfish 26-2 is inserted into a seaweed 26-1 and gaps 26-3 and 26-4 are captured using a deep learning method from multiple images is performed in the same manner as in the first embodiment. Instead of the "meat/bone discrimination" in the first embodiment, the mozuku seaweed 26-1 and the shellfish 26-2 are discriminated according to [0065] to [0113]. Instead of the ultraviolet light irradiation unit 111 constituting the ultraviolet light imaging unit, the purple light irradiation unit 112 constituting the purple light imaging unit, and the blue light irradiation unit 113 constituting the blue light imaging unit constituting the first embodiment, a visible light imaging unit and an ultraviolet light imaging unit are provided.

An eighth embodiment will be described below with reference to Figures 28 and 29. In the eighth embodiment, a different kind of seaweed 28-2 is contained in a mozuku seaweed 28-1. The different kind of seaweed 28-2 emits fluorescence.
(1) In the state shown in FIG. 28 of the eighth embodiment, visible light (wavelength 400 nm to 700 nm) and ultraviolet light (wavelength 10 to 400 nm) are irradiated in combination.
Different species of seaweed 28-2 is visible.
(2) In the state shown in FIG. 29 of the eighth embodiment, ultraviolet light (wavelength 10-400 nm) is irradiated.
Different species of seaweed 28-2 appears faint.
When only visible light is irradiated, the different seaweed 28-2 does not easily glow.

In the eighth embodiment, an image is obtained by superimposing images received using visible light and ultraviolet light, and multiple images received using ultraviolet light, as shown in Figures 28 and 29. These images are superimposed and identified by shape using AI.
The image analysis process executed by the image analysis device 1 using the image discrimination processing unit 14 that has learned feature amounts of a state in which a different type of seaweed 28-2 is included in a seaweed type mozuku 28-1 using a deep learning method from a plurality of images is performed in the same manner as in the first embodiment. Instead of the "meat/bone discrimination" in the first embodiment, discrimination between the seaweed type mozuku 28-1 and the different type of seaweed 28-2 is performed according to [0065] to [0113]. Instead of the ultraviolet light irradiation unit 111 constituting the ultraviolet light imaging unit, the purple light irradiation unit 112 constituting the purple light imaging unit, and the blue light irradiation unit 113 constituting the blue light imaging unit constituting the first embodiment, a visible light imaging unit and an ultraviolet light imaging unit are provided.

A ninth embodiment will be described below with reference to Figures 30 and 31. In the ninth embodiment, a bean sprout 30-1, which is a root vegetable, contains a rotten root portion 30-2. The rotten root portion 30-2 does not emit fluorescence.
(1) In the state shown in FIG. 30 of the ninth embodiment, visible light (wavelength 400 nm to 700 nm) is irradiated.
Root rot area 30-2 is visible.
(2) In the state shown in Fig. 31 of the ninth embodiment, visible light (only red) (wavelength 620-750 nm) is irradiated. Since there is no fluorescent material, red visible light is irradiated.
The root rot area 30-2 appears thin.

In the ninth embodiment, an image is obtained by superimposing an image received using visible light and a plurality of images received using visible light (only red) as shown in Figures 30 and 31. These images are superimposed and identified by shape using AI.
The image analysis process performed by the image analysis device 1 using the image discrimination processing unit 14 that has learned the feature amount of the state in which the root rot part 30-2 is included in the bean sprout 30-1, which is a root vegetable, from a plurality of images using a deep learning method is performed in the same manner as in the first embodiment. Instead of the "meat/bone discrimination" in the first embodiment, discrimination between the bean sprout 30-1, which is a root vegetable, and the root rot part 30-2 is performed according to [0065] to [0113]. Instead of the ultraviolet light irradiation unit 111 constituting the ultraviolet light imaging unit, the purple light irradiation unit 112 constituting the purple light imaging unit, and the blue light irradiation unit 113 constituting the blue light imaging unit constituting the first embodiment, a visible light imaging unit and a visible light (only red) imaging unit are provided.

A tenth embodiment will be described with reference to Figures 32 and 33. In the tenth embodiment, a bean sprout 32-1, which is a root vegetable, contains a blackened damaged portion 32-2. There is no fluorescence emission from the damaged portion 32-2.
(1) In the state shown in FIG. 32 of the tenth embodiment, visible light (wavelength 400 nm to 700 nm) is irradiated.
The damaged area 32-2 is visible. 32-3 is the stem.
(2) In the state shown in Fig. 33 of the tenth embodiment, visible light (only red) (wavelength 620-750 nm) is irradiated. Since there is nothing that emits fluorescence, red visible light is irradiated.
The blackened damaged area 32-2 appears faint.

In the tenth embodiment, an image is obtained by superimposing an image received using visible light and multiple images received using visible light (only red) as shown in Figures 32 and 33. These images are superimposed and identified by shape using AI.
The image analysis process performed by the image analysis device 1 using the image discrimination processing unit 14, which has learned the feature amount of a state in which a blackened damaged part 32-2 is included in a bean sprout 32-1, which is a root vegetable, and a stem 32-3 is also captured, using a deep learning method from multiple images, is performed in the same manner as in the first embodiment. Instead of the "meat/bone discrimination" in the first embodiment, discrimination between the bean sprout 32-1, which is a root vegetable, and the blackened damaged part 32-2 is performed according to [0065] to [0113]. Instead of the ultraviolet light irradiation unit 111 constituting the ultraviolet light imaging unit, the purple light irradiation unit 112 constituting the purple light imaging unit, and the blue light irradiation unit 113 constituting the blue light imaging unit constituting the first embodiment, a visible light imaging unit and a visible light (only red) imaging unit are provided.

An eleventh embodiment will be described below, as shown in Figures 34 and 35. In the eleventh embodiment, a bean sprout 34-1, which is a root vegetable, contains a blackened damaged portion 34-2. The blackened damaged portion 34-2 does not emit fluorescence.
(1) In the state shown in FIG. 34 of the eleventh embodiment, visible light (wavelength 400 nm to 700 nm) is irradiated.
The damaged area 34-2 is visible. 34-3 is the stem.
(2) In the state shown in Fig. 35 of the eleventh embodiment, visible light (only red) (wavelength 620-750 nm) is irradiated. Since there is nothing that emits fluorescence, red visible light is irradiated.
The damaged area 34-2 that has turned black appears faint, while the stem 34-3 appears dim.

In the eleventh embodiment, an image is obtained by superimposing an image received using visible light and multiple images received using visible light (only red) as shown in Figures 34 and 35. These images are superimposed and identified by shape using AI.
The image analysis process performed by the image analysis device 1 using the image discrimination processing unit 14 that has learned feature amounts of a state in which a blackened damaged area 34-2 is included in a bean sprout 34-1, which is a root vegetable, and a stem 34-3 is captured using a deep learning method from multiple images is performed in the same manner as in the first embodiment. Instead of the "meat/bone discrimination" in the first embodiment, discrimination between the bean sprout 34-1, which is a root vegetable, and the blackened damaged area 34-2 is performed according to [0065] to [0113]. Instead of the ultraviolet light irradiation unit 111 constituting the ultraviolet light imaging unit, the purple light irradiation unit 112 constituting the purple light imaging unit, and the blue light irradiation unit 113 constituting the blue light imaging unit constituting the first embodiment, a visible light imaging unit and a visible light (only red) imaging unit are provided.

A twelfth embodiment will be described below, as shown in Figures 36, 37, 38, 39, 40, and 41. In the twelfth embodiment, carbon fishing nets 36-2 and 36-3 are contained in dried seaweed 36-1. The fishing nets 36-2 and 36-3 do not emit fluorescent light.
(1) In the state shown in Figs. 36 to 41 of the twelfth embodiment, near-infrared rays (wavelength 0.7 to 2.5 µm) are irradiated.
Carbon fishing net 36-2 is visible.
(2) In the twelfth embodiment shown in Figures 36, 37, 39, 40, and 41, near-infrared light was applied from the side. Since near-infrared light was applied from the side, the side surface appeared white and was easy to see.

In the twelfth embodiment, an image is obtained by superimposing a plurality of images received using near-infrared light or visible light other than near-infrared light, as shown in Figures 36 to 41. These images are superimposed and identified by shape using AI.
The image analysis process performed by the image analysis device 1 using the image identification processing unit 14 that has learned the feature amount of the state in which the carbon fishing nets 36-2 and 36-3 are contained in the dried seaweed 36-1, which is dried seaweed, using a deep learning method from multiple images, is performed in the same manner as in the first embodiment. Instead of the "meat bone discrimination" in the first embodiment, the discrimination between the dried seaweed 36-1 and the carbon fishing nets 36-2 and 36-3 is performed according to [0065] to [0113]. Instead of the ultraviolet light irradiation unit 111 constituting the ultraviolet light imaging unit, the purple light irradiation unit 112 constituting the purple light imaging unit, and the blue light irradiation unit 113 constituting the blue light imaging unit constituting the first embodiment, a near-infrared light imaging unit and a visible light imaging unit other than near-infrared light are provided.

A thirteenth embodiment will be described below, as shown in Figures 42, 43, 44, and 45. In the thirteenth embodiment, a carbon fishing net 42-2 is placed in a dried wakame seaweed 42-1, which is a type of dried seaweed. The carbon fishing net 42-2 does not emit fluorescent light.
(1) In the state shown in Figs. 36 to 41 of the thirteenth embodiment, near-infrared light (wavelength 0.7 to 2.5 µm) is irradiated.
The carbon fishing net 42-2 is visible. Since the near-infrared light is irradiated from the side 42-3, the side 42-3 appears white and is easy to see.
(2) In the examples shown in Figs. 43, 44 and 45, near-infrared light was applied from the side.

In the thirteenth embodiment, an image is obtained by superimposing a plurality of images received using near-infrared light or visible light as shown in Figures 42 to 45. These images are superimposed and identified by shape using AI.
The image analysis process executed by the image analysis device 1 using the image identification processing unit 14 that has learned the feature amount of the state in which the carbon fishing net 42-2 is contained in the dried wakame 42-1, which is dried seaweed, from multiple images using a deep learning method is processed in the same manner as in the first embodiment. Instead of the "meat bone discrimination" in the first embodiment, the discrimination between the dried wakame 42-1, which is dried seaweed, and the carbon fishing net 42-2 is performed according to [0065] to [0113]. Instead of the ultraviolet light irradiation unit 111 constituting the ultraviolet light imaging unit, the purple light irradiation unit 112 constituting the purple light imaging unit, and the blue light irradiation unit 113 constituting the blue light imaging unit constituting the first embodiment, a near-infrared light imaging unit and a visible light imaging unit other than near-infrared light are provided.

A fourteenth embodiment will be described below, as shown in Figures 46, 47, 48, and 49. In the fourteenth embodiment, tiny flying insects 46-2, 46-3, 46-4, and 46-5 are contained in a soba noodle 46-1. There is no fluorescence.
(1) In the state shown in Figs. 46 to 49 of the fourteenth embodiment, near-infrared light (wavelength 0.7 to 2.5 µm) is irradiated.
Tiny flying insect 46-2 is visible.
In the fourteenth embodiment shown in Figures 46 to 49, near-infrared light was applied from the side. Since near-infrared light was applied from the side, the side surface appeared white and was easy to see.
(2) White light was applied (not shown).

In the fourteenth embodiment, an image is obtained by superimposing a plurality of images received using near-infrared light or visible light and white light as shown in Figures 46 to 49. These images are superimposed and identified by shape using AI.
The image analysis process performed by the image analysis device 1 using the image discrimination processing unit 14 that has learned the feature amount of the state in which the tiny flying insect 46-2 is in the buckwheat noodles 46-1 using a deep learning method from multiple images is performed in the same manner as in the first embodiment. Instead of the "meat/bone discrimination" in the first embodiment, the discrimination between the buckwheat noodles 46-1 and the tiny flying insects 46-2, 46-3, 46-4, and 46-5 is performed according to [0065] to [0113]. Instead of the ultraviolet light irradiation unit 111 constituting the ultraviolet light imaging unit, the purple light irradiation unit 112 constituting the purple light imaging unit, and the blue light irradiation unit 113 constituting the blue light imaging unit constituting the first embodiment, a near-infrared light imaging unit and a visible light imaging unit other than near-infrared light are provided.

A fifteenth embodiment shown in Figures 50 and 51 will be described. In the fifteenth embodiment, a transparent vinyl 50-2 is placed inside a transparent vinyl-encased pork chop 50-1. The transparent vinyl 50-2 does not emit fluorescent light. It does have diffuse reflection. 50-4 is a transparent vinyl bag.
(1) In the state shown in Fig. 50 of the fifteenth embodiment, visible light (wavelength 400 nm to 700 nm) is irradiated. There is diffuse reflection 50-3 from a transparent vinyl bag 50-4.
(2) In the state shown in Fig. 51 of the fifteenth embodiment, visible light (only red) (wavelength 620-750 nm) is irradiated. Since there is nothing that emits fluorescence, red visible light is irradiated.

In the fifteenth embodiment, an image is obtained by superimposing an image received using visible light and multiple images received using visible light (only red) as shown in Figures 50 and 51. These images are superimposed and identified by shape using AI.
The image analysis process performed by the image analysis device 1 using the image discrimination processing unit 14 that has learned the feature amount of the transparent vinyl-encased pork pieces 50-1 with the transparent vinyl 50-2 encased therein from a plurality of images using a deep learning technique is performed in the same manner as in the first embodiment. Instead of the "meat bone discrimination" in the first embodiment, the transparent vinyl-encased pork pieces 50-1 and the transparent vinyl 50-2 are discriminated according to [0065] to [0113]. Instead of the ultraviolet light irradiation unit 111 constituting the ultraviolet light imaging unit, the purple light irradiation unit 112 constituting the purple light imaging unit, and the blue light irradiation unit 113 constituting the blue light imaging unit constituting the first embodiment, a visible light imaging unit and a visible light (only red) imaging unit are provided.

A 16th embodiment shown in Figures 52 and 53 will be described. In the 16th embodiment, a transparent vinyl 52-2 is placed inside a transparent vinyl-encased pork chop 52-1. The transparent vinyl 52-2 does not emit fluorescent light. There is diffuse reflection. 52-4 is a transparent vinyl bag.
(1) In the state shown in Fig. 52 of the 16th embodiment, visible light (wavelength 400 nm to 700 nm) is irradiated. There is diffuse reflection 52-3 from a transparent vinyl bag 52-4.
(2) In the state shown in Fig. 53 of the 16th embodiment, visible light (only red) (wavelength 620-750 nm) is irradiated. Since there is nothing that emits fluorescence, red visible light is irradiated.

In the 16th embodiment, an image is obtained by superimposing an image received using visible light and a plurality of images received using visible light (only red) as shown in Figures 52 and 53. These images are superimposed and identified by shape using AI.
The image analysis process executed by the image analysis device 1 using the image discrimination processing unit 14 that has learned the feature amount of the transparent vinyl-encased pork pieces 52-1 in a state where the transparent vinyl is contained from a plurality of images using a deep learning method is performed in the same manner as in the first embodiment. Instead of the "meat bone discrimination" in the first embodiment, the transparent vinyl-encased pork pieces 52-1 and the transparent vinyl 52-2 are discriminated according to [0065] to [0113]. Instead of the ultraviolet light irradiation unit 111 constituting the ultraviolet light imaging unit, the purple light irradiation unit 112 constituting the purple light imaging unit, and the blue light irradiation unit 113 constituting the blue light imaging unit constituting the first embodiment, a visible light imaging unit and a visible light (only red) imaging unit are provided.

A seventeenth embodiment shown in Figures 54 and 55 will be described. In the seventeenth embodiment, a transparent vinyl 54-2 is placed inside a transparent vinyl-encased pork chop 54-1. The transparent vinyl 54-2 does not emit fluorescent light. It does have diffuse reflection. 54-4 is a transparent vinyl bag.
(1) In the state shown in Fig. 52 of the 17th embodiment, visible light (wavelength 400 nm to 700 nm) is irradiated. There is diffuse reflection 54-3 from a transparent vinyl bag 54-4.
(2) In the state shown in Fig. 55 of the 17th embodiment, visible light (only red) (wavelength 620-750 nm) is irradiated. Since there is nothing that emits fluorescence, red visible light is irradiated.

In the seventeenth embodiment, an image is obtained by superimposing an image received using visible light and multiple images received using visible light (only red) as shown in Figures 54 and 55. These images are superimposed and identified by shape using AI.
The image analysis process executed by the image analysis device 1 using the image discrimination processing unit 14 that has learned the feature amount of the transparent vinyl-encased pork pieces 54-1 in a state where the transparent vinyl is contained from a plurality of images using a deep learning method is performed in the same manner as in the first embodiment. Instead of the "meat bone discrimination" in the first embodiment, the transparent vinyl-encased pork pieces 54-1 and the transparent vinyl 54-2 are discriminated according to [0065] to [0113]. Instead of the ultraviolet light irradiation unit 111 constituting the ultraviolet light imaging unit, the purple light irradiation unit 112 constituting the purple light imaging unit, and the blue light irradiation unit 113 constituting the blue light imaging unit constituting the first embodiment, a visible light imaging unit and a visible light (only red) imaging unit are provided.

An 18th embodiment will be described below with reference to Figures 56 to 59. The 18th embodiment is a croquette 56-1 which is a deep-fried food containing peeled off batter or frozen parts 56-2.
In the state shown in FIG. 56 of the 18th embodiment, a fried food, croquette 56-1, having peeled off batter or frozen portion 56-2, was irradiated with visible light (only blue light) (wavelength 450-495 nm).
In the state shown in FIG. 57 of the 18th embodiment, a fried food, croquette 56-1 having peeled off batter or frozen portion 56-2, was irradiated with visible light (only red light) (wavelength 620-750 nm).
In the state shown in FIG. 58 of the 18th embodiment, visible light (white light) was irradiated onto a fried food, croquette 56-1, having peeled off batter or frozen portion 56-2.
In the state shown in FIG. 59 of the 18th embodiment, a fried food, croquette 56-1, having peeled off batter or frozen portion 56-2, was irradiated with visible light (only green light) (wavelength 495-570 nm).
In the case of croquette 56-1 (potato croquette), the peeling of the coating or the frozen portion 56-2 can be clearly confirmed by using white light plus red light.

In the 18th embodiment, an image is obtained by superimposing a plurality of images received using visible light as shown in Figures 56 to 59. These images are superimposed and identified by shape using AI.
The image analysis process executed by the image analysis device 1 using the image recognition processing unit 14, which has learned the features of the state of a croquette 56-1, a fried food containing peeled off batter or frozen parts 56-2, from multiple images using a deep learning technique, is performed in the same manner as in the first embodiment.
Instead of the "meat or bone discrimination" in the first embodiment, discrimination between croquette 56-1, which is a fried food, and peeling or frozen parts 56-2 of the fried food batter is performed according to [0065] to [0113]. Instead of ultraviolet light irradiating section 111 constituting the ultraviolet light imaging section, purple light irradiating section 112 constituting the purple light imaging section, and blue light irradiating section 113 constituting the blue light imaging section constituting the first embodiment, a visible light (only blue light) imaging section, a visible light (only red light) imaging section, a visible light (white light) imaging section, and a visible light (only green light) imaging section are provided.

A description will be given of a 19th embodiment shown in Figures 60 to 63. The 19th embodiment is a cream croquette 60-1 which is a fried food containing peeled off batter or frozen parts 60-2.
In the state shown in FIG. 60 of the 19th embodiment, a cream croquette 60-1, which is a fried food having peeled off batter or frozen portion 60-2, is irradiated with visible light (only blue light) (wavelength 450-495 nm).
In the state shown in FIG. 61 of the 19th embodiment, a cream croquette 60-1, which is a fried food having peeled off batter or a frozen portion 60-2, is irradiated with visible light (only red light) (wavelength 620-750 nm).

In the state shown in FIG. 62 of the 19th embodiment, visible light (white light) was irradiated onto a deep-fried cream croquette 60-1 having peeled or frozen portions 60-2 of the batter of the deep-fried food.
In the state shown in FIG. 63 of the 19th embodiment, a cream croquette 60-1, which is a fried food having peeled off batter or a frozen portion 60-2, is irradiated with visible light (only green light) (wavelength 495-570 nm).
In the case of the cream croquette 60-1, the peeling of the coating or the frozen portion 60-2 can be clearly seen with white light plus red light.

In the nineteenth embodiment, an image is obtained by superimposing a plurality of images received using visible light as shown in Figures 60 to 63. These images are superimposed and identified by shape using AI.
The image analysis process performed by the image analysis device 1 using the image identification processing unit 14 that has learned the feature amount of the state of the cream croquette 60-1, which is a fried food mixed with peeled off batter or frozen part 60-2, from multiple images using a deep learning method is performed in the same manner as in the first embodiment. Instead of the "meat bone discrimination" in the first embodiment, the discrimination between the cream croquette 60-1, which is a fried food, and the peeled off batter or frozen part 60-2 of the fried food is performed according to [0065] to [0113]. Instead of the ultraviolet light irradiation unit 111 constituting the ultraviolet light imaging unit, the purple light irradiation unit 112 constituting the purple light imaging unit, and the blue light irradiation unit 113 constituting the blue light imaging unit constituting the first embodiment, a visible light (only blue light) imaging unit, a visible light (only red light) imaging unit, a visible light (white light) imaging unit, and a visible light (only green light) imaging unit are provided.

A twentieth embodiment will be described below with reference to Figures 64 to 67. The twentieth embodiment is fried chicken 64-1, which is a fried food containing peeled or frozen portions 64-2 of the fried food.
In the state shown in FIG. 64 of the twentieth embodiment, a fried chicken 64-1 having peeled or frozen portions 64-2 of the batter was irradiated with visible light (only blue light) (wavelengths 450-495 nm).
In the state shown in FIG. 65 of the twentieth embodiment, a fried chicken 64-1 having peeled or frozen portions 64-2 of the batter was irradiated with visible light (only red light) (wavelength 620-750 nm).

In the state shown in FIG. 66 of the twentieth embodiment, a fried chicken 64-1 having peeled or frozen portions 64-2 of the batter was irradiated with visible light (only green light) (wavelength 495-570 nm).
In the state shown in FIG. 67 of the twentieth embodiment, a piece of fried chicken 64-1 having peeled or frozen portions 64-2 of the batter was irradiated with visible light (blue plus white).
In the case of fried chicken 64-1, the peeling of the coating or the frozen parts 64-2 can be clearly seen by using white light plus red light.

In the twentieth embodiment, an image is obtained by superimposing a plurality of images received using visible light as shown in Figures 64 to 67. These images are superimposed and identified by shape using AI.
The image analysis process performed by the image analysis device 1 using the image identification processing unit 14 that has learned the feature amount of the state of fried chicken 64-1, which is a fried food mixed with peeled off batter or frozen part 64-2, from multiple images using a deep learning method is performed in the same manner as in the first embodiment. Instead of the "meat bone discrimination" in the first embodiment, discrimination between cream croquette 60-1, which is a fried food, and peeled off batter or frozen part 64-2 of the fried food is performed according to [0065] to [0113]. Instead of the ultraviolet light irradiation unit 111 constituting the ultraviolet light imaging unit, the purple light irradiation unit 112 constituting the purple light imaging unit, and the blue light irradiation unit 113 constituting the blue light imaging unit constituting the first embodiment, a visible light (only blue light) imaging unit, a visible light (only red light) imaging unit, a visible light (only green light) imaging unit, and a visible light (blue plus white light) imaging unit are provided.

When detecting the peeled-off portion of a croquette, even though cream croquette and potato croquette are both deep-fried foods, the color that reacts to the peeled off portion is white, i.e., visible light, and another visible color.
For cream croquettes, the peeled off parts of the batter can be clearly seen with white + green. For potato croquettes, the peeled off parts of the batter can be clearly seen with white + red. For fried chicken, the peeled off parts of the batter can be clearly seen with white + blue.
In this way, by shining light that can extract characteristics when searching for peeled parts of fried foods and having the AI learn the two resulting screens, the accuracy of detecting peeled parts of fried foods can be improved.

REFERENCE SIGNS LIST 1 Image analysis device 2 Input unit 3 Output unit 11 Irradiation unit 111 Ultraviolet light irradiation unit 112 Purple light irradiation unit 113 Blue light irradiation unit 12 Operation input information acquisition unit 13 Image data division processing unit 14 Image identification processing unit 15 Identification result conversion processing unit 16 Image processing unit 17 Output processing unit 18 Storage unit 21 Image data acquisition unit
A Bone B Bone C Bone D Bone E Cartilage F Muscle fiber bundle G Endomysium H Perimysium 16-1 Mozuku seaweed 16-2 Fishing line (yellow)
18-1 Mozuku seaweed 18-2 Fishing line (green)
20-1 Mozuku seaweed 20-2 Fishing line (brown)
22-1 Mozuku seaweed 22-2 Shrimp 24-1 Mozuku seaweed 24-2 Fish 26-1 Mozuku seaweed 26-2 Shellfish 28-1 Mozuku seaweed 28-2 Different seaweed 30-1 Bean sprouts 30-2 Root rot 32-1 Bean sprouts 32-2 Blackened damaged area Bean sprouts 34-1 Bean sprouts 34-2 Blackened damaged area 36-1 Dried seaweed 36-2 Carbon fishing net 36-3 Carbon fishing net 42-1 Dried wakame seaweed 42-2 Carbon fishing net 46-1 Soba noodles 46-2 Tiny flying insects 46-3 Tiny flying insects 46-4 Tiny flying insects 46-5 Tiny flying insects 50-1 Pork shreds in clear vinyl 50-2 Clear vinyl 50-3 Diffuse reflection 50-4 Clear vinyl bag 52-1 Pork shreds in clear plastic 52-2 Clear plastic 52-3 Diffuse reflection 52-4 Clear plastic bag 54-1 Pork shreds in clear plastic 54-2 Clear plastic 54-3 Diffuse reflection 54-4 Clear plastic bag 56-1 Deep-fried croquette 56-2 Peeling or frozen part of deep-fried batter 60-1 Cream croquette 60-2 Peeling or frozen part of deep-fried batter 64-1 Fried chicken 64-2 Peeling or frozen part of deep-fried batter

Claims (8)

In the dark,
an ultraviolet light photographing unit that irradiates ultraviolet light onto a cut surface of meat and obtains an ultraviolet image by collecting reflected light of the ultraviolet light;
a purple light photographing unit that irradiates a cut surface of meat with purple light and obtains a purple light image by collecting reflected light of the purple light;
a blue light photographing unit that irradiates blue light onto a cut surface of meat and captures reflected light of the blue light to obtain a blue light image;
A meat bone discrimination device that discriminates a bone part of a cut surface of meat when extracting a bone , which is an extraction target, from image data obtained from the above,
A partitioning means for partitioning the image data;
an image classification means for calculating a probability that each of the segments of the image data divided by the segment division means is the object to be extracted based on the machine learning result;
a classification result conversion means for converting the probability that the image is the extraction target for each of the categories, which is calculated by the image classification means, into a pixel value;
an image processing means for performing image processing on an image having the pixel values converted by the classification result conversion means ,
This meat bone discriminating device is characterized in that when irradiated with ultraviolet light, a part that appears white is judged to be a bone, when irradiated with purple visible light, a part that appears whitish yellow is judged to be a bone, and when irradiated with blue visible light, a part that appears yellow is judged to be a bone . The image processing means performs image processing on the image data divided by the division means,
2. The meat bone discrimination device according to claim 1, wherein the image recognition means calculates a probability that the object to be extracted is a bone for each of the image data sections processed by the image processing means. 3. The meat bone discriminating apparatus according to claim 1, wherein an image is obtained by passing an orange or yellow filter through the ultraviolet light photographing section. 4. The meat bone discriminating device according to claim 1, 2 or 3, wherein the image is obtained by passing an orange or yellow filter through the blue light photographing section. In the dark,
an ultraviolet light photographing unit that irradiates ultraviolet light onto a cut surface of meat and obtains an ultraviolet image by collecting reflected light of the ultraviolet light;
a purple light photographing unit that irradiates a cut surface of meat with purple light and obtains a purple light image by collecting reflected light of the purple light;
a blue light photographing unit that irradiates blue light onto a cut surface of meat and captures reflected light of the blue light to obtain a blue light image;
obtained from
A meat bone discrimination method for discriminating a bone portion of a cut surface of meat when extracting a bone as an extraction target from image data, comprising the steps of:
A partitioning step of partitioning the image data;
an image classification step of calculating, for each of the segments of the image data divided by the processing of the segment division step, a probability that the segment is the object to be extracted, based on a machine learning result of the feature amount of the object to be extracted by a deep learning technique;
a classification result conversion step of converting the probability of the object being the extraction target for each of the categories, calculated by the processing of the image classification step, into a pixel value;
an image processing step of performing image processing on an image having pixel values converted by the processing of the classification result conversion step;
a determining step of determining that a portion that appears white is a bone when irradiated with ultraviolet light, that a portion that appears whitish-yellow is a bone when irradiated with purple visible light, and that a portion that appears yellow is a bone when irradiated with blue visible light;
A method for identifying meat bones, comprising: 6. The method for identifying bones in meat according to claim 5, wherein the image is obtained by passing an orange or yellow filter through the purple light photographing section. 7. The method for identifying meat bones according to claim 5 or 6, wherein the image is obtained by passing an orange or yellow filter through the blue light photographing section. In the dark,
an ultraviolet light photographing unit that irradiates ultraviolet light onto a cut surface of meat and obtains an ultraviolet image by collecting reflected light of the ultraviolet light;
a purple light photographing unit that irradiates a cut surface of meat with purple light and obtains a purple light image by collecting reflected light of the purple light;
a blue light photographing unit that irradiates blue light onto a cut surface of meat and captures reflected light of the blue light to obtain a blue light image;
obtained from
A program for causing a computer to execute a meat bone discrimination process for discriminating a bone portion of a cut surface of meat when extracting a bone as an extraction target from image data, comprising:
A partitioning step of partitioning the image data;
an image classification step of calculating, for each of the segments of the image data divided by the processing of the segment division step, a probability that the segment is the object to be extracted, based on a machine learning result of the feature amount of the object to be extracted by a deep learning technique;
a classification result conversion step of converting the probability of the object being the extraction target for each of the categories, calculated by the processing of the image classification step, into a pixel value;
an image processing step of performing image processing on an image having the pixel values converted by the processing in the classification result conversion step;
a determining step of determining that a portion that appears white is a bone when irradiated with ultraviolet light, that a portion that appears whitish-yellow is a bone when irradiated with purple visible light, and that a portion that appears yellow is a bone when irradiated with blue visible light;
A program comprising:

Images