KR20230137594A - Method and apparatus for determining fish grade using artificial neural network - Google Patents

Abstract

The present invention provides a fish grade determination method in which, when an image of a cross-section of a fish is input, the image is pre-processed, and the pre-processed image is input to a pre-trained fish grade determination neural network to determine the fish grade.

Description

Fish grading method and device using artificial neural network {METHOD AND APPARATUS FOR DETERMINING FISH GRADE USING ARTIFICIAL NEURAL NETWORK}

The present invention relates to a method and device for determining the grade of fish using an artificial neural network, and more specifically, to a method and device for determining the grade of fish using an artificial neural network.

Most of the tuna brought into Korea is caught from deep-sea fishing boats operating in the South Pacific and Indian Oceans, and the freshness of the tuna changes significantly depending on the processing process that removes intestines, blood, and gills immediately after catching. Afterwards, it is quickly frozen at an ultra-low temperature of -55 degrees Celsius or lower and transported domestically. In the case of tuna products for sashimi, thawing is not performed until it is shipped through the production process after warehousing.

Accordingly, in the case of tuna products for sashimi, only a small portion of the fish body is cut (e.g., tail section within 5 mm), and individual grades are selected after thawing. Because this process involves somewhat specialized parts, such as observing protein denaturation, the work relies on a small number of skilled experts, and their dependency is evaluated higher than in other industries.

Meanwhile, the aging of experts continues to progress, and furthermore, the process of nurturing experts takes a considerable amount of time and money, limiting the training of successors. In addition, even if an expert judges the grade, there may be a lack of objectivity in ambiguous judgments, and the lack of standardized standards can serve as a reason for disputes between transactions.

Domestic Registered Patent No. 10-1926490 (2018.12.03.) Domestic Registered Patent No. 10-2337383 (2021.12.06.)

The technical problem to be solved by the present invention is to provide a method and device for determining a grade by inputting an image of a fish into a pre-trained artificial neural network.

In addition, another technical problem to be solved by the present invention is to provide a method and device for training an artificial neural network to determine the grade of fish.

One aspect of the present invention includes, when an image of a cross-section of a fish is input, preprocessing the image; and inputting the pre-processed image into a pre-trained fish grade determination neural network to determine the grade of the fish.

In addition, the fish grade determination neural network is learned according to an artificial neural network learning method, and the artificial neural network learning method includes receiving a learning image for a fish and a correct grade as label data for the learning image; and training the fish grade determination neural network to output a learning grade according to the learning image using the learning image and the correct answer grade.

Additionally, the image may be one of a tail cut surface image obtained by photographing a cut surface of the tail of the fish and a midbone cut surface image obtained by photographing a cut surface of the midbone of the fish.

In addition, the preprocessing step may include correcting the color value of a pixel corresponding to the bone of the fish among a plurality of pixels included in the image with the color value of a pixel corresponding to the flesh of the fish. .

In addition, the step of correcting the color value of the pixel corresponding to the fish flesh includes classifying the plurality of pixels into a plurality of color groups; Identifying a first color group corresponding to the fish bone among the plurality of color groups; identifying a second color group with the largest number of pixels among the plurality of color groups; and replacing the color value of each of at least one pixel included in the first color group using the color value of at least one pixel included in the second color group.

Additionally, the preprocessing step may include, when multiple images are input, matching them by assigning different weights to each of the multiple images.

In addition, the step of matching by applying different weights is such that the color of the cross section of the fish is emphasized in the first image among the plurality of images, and the boundary line between the meat and fat of the fish is emphasized in the second image. Weights may be assigned to each of the first image and the second image.

Additionally, the plurality of images may include a first image captured in a first lighting environment and a second image captured in a second lighting environment with lower brightness than the first lighting environment.

Another aspect of the present invention includes receiving a learning image for a fish and a correct answer grade as label data for the learning image; And it may include the step of training a fish grade determination neural network to output a learning grade according to the learning image using the learning image and the correct answer grade.

In addition, the learning image may be one of a learning tail cut surface image obtained by photographing a tail cut surface of the fish and a learning midbone cut surface image obtained by photographing a midbone cut surface of the fish.

In addition, the step of training the fish grade determination neural network includes correcting the color value of a pixel corresponding to the bone of the fish among a plurality of pixels included in the learning image with the color value of a pixel corresponding to the flesh of the fish. May include ;.

In addition, the step of training the fish grade determination neural network may include, when a plurality of training images are input, matching them by assigning different weights to each of the plurality of training images.

Another aspect of the present invention is a memory in which a pre-trained fish grading neural network is stored; And when an image of a cross-section of a fish is input, a processor that pre-processes the image and inputs the pre-processed image to the fish grade determination neural network to determine the grade of the fish.

Another aspect of the present invention is a computer-readable recording medium storing a computer program, wherein the computer program, when executed by a processor, preprocesses the image when an image of a cross-section of a fish is input; and determining the grade of the fish by inputting the pre-processed image into a pre-trained fish grade determination neural network. It may include instructions for causing the processor to perform the method including.

Another aspect of the present invention is a computer program stored in a computer-readable recording medium, which, when executed by a processor, preprocesses the image when an image of a cross-section of a fish is input. ; and determining the grade of the fish by inputting the pre-processed image into a pre-trained fish grade determination neural network. It may include instructions for causing the processor to perform the method including.

Another aspect of the present invention is a memory that stores a learning image for a fish and a correct answer grade as label data for the learning image; and a processor that trains a fish rating neural network to output a learning rating according to the learning image using the learning image and the correct answer rating.

Another aspect of the present invention is a computer-readable recording medium storing a computer program, which, when executed by a processor, produces a learning image for a fish and a correct answer rating as label data for the learning image. Step of receiving input; and training a fish rating neural network to output a learning rating according to the learning image using the learning image and the correct answer rating. It may include instructions for causing the processor to perform the method including.

Another aspect of the present invention is a computer program stored in a computer-readable recording medium, which, when executed by a processor, provides a learning image for fish and a correct answer grade as label data for the learning image. Step of receiving input; and training a fish rating neural network to output a learning rating according to the learning image using the learning image and the correct answer rating. It may include instructions for causing the processor to perform the method including.

According to one aspect of the present invention described above, by providing a fish grade determination method and device using an artificial neural network, the fish grade can be determined by inputting an image of a fish into a pre-trained artificial neural network.

In addition, according to another aspect of the present invention described above, by providing an artificial neural network learning method and device, an artificial neural network can be trained to determine the grade of fish through images of fish.

1 is a block diagram of an artificial neural network learning device according to an embodiment of the present invention.
Figure 2 is a block diagram of a fish grading device according to an embodiment of the present invention.
Figure 3 is a block diagram conceptually showing an embodiment of the fish grade determination program of Figure 2.
Figures 4 and 5 are block diagrams showing an example of a process for determining the grade of fish in the grade determination unit of Figure 3.
Figure 6 is a flowchart of an artificial neural network learning method according to an embodiment of the present invention.
Figure 7 is a flowchart of a fish grading method according to an embodiment of the present invention.
Figures 8 and 9 are detailed flowcharts of preprocessing the image of Figure 7.

The advantages and features of the present invention and methods for achieving them will become clear by referring to the embodiments described in detail below along with the accompanying drawings. However, the present invention is not limited to the embodiments disclosed below and may be implemented in various different forms. The present embodiments are merely provided to ensure that the disclosure of the present invention is complete and to be understood by those skilled in the art. It is provided to fully inform those who have the scope of the invention, and the present invention is only defined by the scope of the claims.

In describing embodiments of the present invention, if a detailed description of a known function or configuration is judged to unnecessarily obscure the gist of the present invention, the detailed description will be omitted. The terms described below are terms defined in consideration of functions in the embodiments of the present invention, and may vary depending on the intention or custom of the user or operator. Therefore, the definition should be made based on the contents throughout this specification.

Hereinafter, preferred embodiments of the present invention will be described in detail with reference to the attached drawings.

1 is a block diagram of an artificial neural network learning device according to an embodiment of the present invention.

Referring to FIG. 1, the artificial neural network learning device 100 can train a fish rating neural network to determine the rating of a fish according to an image of the fish.

Here, the image of the fish is an image of a cross-section of the fish, and the grade of the fish may be a classification of freshness, quality, etc. of the fish.

In one embodiment, the image of the fish may be one of a tail cut surface image obtained by photographing a cut surface of the tail of a fish and a midbone cut surface image obtained by photographing a cut surface of the midbone of a fish.

Accordingly, the artificial neural network learning device 100 includes an image of the cut surface of the fish's tail, a tail cut surface rating model that outputs the grade of the fish when the correct grade is input as label data, a cut surface image of the midbone of the fish, and the correct grade as label data. When this is input, at least one of the midbone cut surface grade determination models that output the grade of fish can be trained.

At this time, the tail cut surface grading model judges the fish to be one of grades A, B, and C according to the tail cut surface image, and the midbone cut surface grading model judges the fish to be one of A, B, C, and D grades according to the midbone cut surface image. It can be learned to judge as one.

To this end, the artificial neural network learning device 100 may include an input/output module 110, a processor 120, and a memory 130.

The input/output module 110 can receive a learning image for a fish and a correct answer grade as label data for the learning image.

In one embodiment, when a tail cut surface image for learning is input, the input/output module 110 inputs any one of A, B, and C grades as the correct grade for the tail cut surface image for learning, and a midbone cut surface image for learning is input. In this case, any one of A, B, C, and D grades can be input as the correct grade for the midbone cross section image for learning.

At this time, the memory 130 may store at least one of a learning image input through the input/output module 110, a correct answer grade, and a fish grade judgment neural network that is being learned or has completed learning.

In this regard, in one embodiment, the correct answer grade may be set based on shrinkage due to thawing stiffening phenomenon, protein denaturation due to heat, and meat color, etc. Additionally, in another embodiment, the fish determined based on the tail cut image may be yellowfin tuna and bigeye tuna, and the fish determined based on the midbone cut image may be a fish species caught through purse seine fishing.

The processor 120 may train a fish rating neural network to output a learning rating according to the learning image using the learning image and the correct answer rating.

To this end, the processor 120 may perform learning according to the learning image and the correct answer grade using an artificial neural network algorithm such as CNN (Convolution Neural Network).

Here, CNN may be an artificial neural network learning technique that extracts a feature map from an image through one or more convolution layers and outputs results according to the extracted feature map.

Meanwhile, the processor 120 may perform preprocessing on the training image before training the fish rating neural network.

For example, the processor 120 may correct the color value of a pixel corresponding to a fish bone among a plurality of pixels included in a training image to the color value of a pixel corresponding to the fish meat.

Additionally, when multiple training images are input, the processor 120 may match them by assigning different weights to each of the multiple training images.

In this regard, the process of performing pre-processing on the image may be the same as or similar to the pre-processing process performed in the fish rating device, and therefore, the process of performing pre-processing on the image will be described in more detail below.

Figure 2 is a block diagram of a fish grading device according to an embodiment of the present invention.

Referring to FIG. 2, when an image of a cross-section of a fish is input, the fish grading device 200 pre-processes the image and inputs the pre-processed image into the pre-trained fish grading neural network 400 to grade the fish. can be judged.

Here, the fish grade determination neural network 400 may be learned by the artificial neural network learning device 100 of FIG. 1.

To this end, the fish grade determination device 200 may include an input/output module 210, a processor 220, and a memory 230.

An image of a cross-section of a fish may be input to the input/output module 210. Accordingly, the image input through the input/output module 210 may be stored in the memory 230 .

At this time, the image taken of the cross section of the fish may be either a tail cut surface image or a midbone cut surface image.

Meanwhile, a pre-trained fish grade determination neural network 400 may be stored in the memory 230. At this time, the fish rating neural network 400 stored in the memory 230 may include a tail cut surface rating model and a midbone cut surface rating model.

In this regard, in one embodiment, the input/output module 210 may receive a command to select one of the tail cut surface grade determination model and the midbone cut surface grade determination model from the user. In this case, the processor 220 may determine the grade of the fish using the fish grade determination neural network 400 according to the user's selection command.

Meanwhile, the memory 230 may store the fish grade determination program 300 and information necessary for execution of the fish grade determination program 300.

In this specification, the fish grading program 300 pre-processes images taken of the cross-section of fish, and inputs the pre-processed image into the pre-trained fish grading neural network 400 to determine the grade of the fish. It can mean software that does.

Accordingly, the processor 220 may load the fish grading program 300 and information necessary for execution of the fish grading determination program 300 from the memory 230 in order to execute the fish grading determination program 300.

Meanwhile, the function and/or operation of the fish grade determination program 300 will be examined in detail with reference to FIG. 3.

Figure 3 is a block diagram conceptually showing an embodiment of the fish grade determination program of Figure 2.

Referring to FIG. 3 , the fish grading program 300 may include a color correction unit 310, an image matching unit 320, and a grading unit 330.

The color correction unit 310, image matching unit 320, and rating determination unit 330 shown in FIG. 3 perform the functions of the fish rating determination program 300 in order to easily explain the functions of the fish rating determination program 300. It is conceptually divided and is not limited to this. Depending on the embodiment, the functions of the color correction unit 310, the image matching unit 320, and the rating determination unit 330 may be merged/separated and may be implemented as a series of instructions included in one program. .

If it is determined that fish bones are captured in the image, the color correction unit 310 may correct the color of the fish bone area. To this end, the color correction unit 310 may correct the color value of a pixel corresponding to the bone of a fish among a plurality of pixels included in the image to the color value of a pixel corresponding to the flesh of the fish.

At this time, the color correction unit 310 may include a classification unit 311 and a processing unit 312.

The classification unit 311 may classify multiple pixels included in the image into multiple color groups. In one embodiment, the classification unit 311 may classify a plurality of pixels into color groups according to each color range based on a plurality of preset color ranges.

At this time, the classification unit 311 may determine a plurality of different color ranges based on the average color value for a plurality of pixels. In one embodiment, as the brightness of the average color value increases, the classification unit 311 may narrow the range for colors with high brightness and set the range for colors with low brightness to widen. Additionally, as the brightness of the average color value decreases, the classification unit 311 may set the range for colors with high brightness to be wider and the range for colors with lower brightness to be narrow.

In another embodiment, as the saturation of the average color value increases, the classification unit 311 may set the range for high-saturation colors to be narrow and the range for low-saturation colors to be wide. Additionally, as the saturation of the average color value decreases, the classification unit 311 may set the range for high-saturation colors to be wider and the range for low-saturation colors to be narrower.

In another embodiment, the classification unit 311 may set the color range to be narrower for colors closer to the saturation of the average color value, and may set the color range wider for colors farther from the saturation of the average color value. Additionally, the classification unit 311 may set the color range narrower for colors closer to the brightness of the average color value, and set the color range wider for colors farther away from the brightness of the average color value.

Accordingly, the processing unit 312 can identify the first color group corresponding to the fish bone among the plurality of color groups and the second color group with the largest number of pixels among the plurality of color groups.

In one embodiment, the first color group may include colors from [242, 226, 226] to [255, 255, 255]. Here, [242, 226, 226] is an RGB color value, which may be a color value in which the R element is set to 242, the G element is set to 226, and the B element is set to 226.

Meanwhile, the second color group is a color with the largest proportion in the area corresponding to fish, and may be, for example, the color of fish flesh.

Accordingly, the processing unit 312 may replace the color value of each of at least one pixel included in the first color group using the color value of at least one pixel included in the second color group. For example, the processing unit 312 may replace the color value of each of at least one pixel included in the first color group with the average color value of at least one pixel included in the second color group.

Meanwhile, the color correction unit 310 may correct the color of the fish bone using only the area corresponding to the fish among the plurality of pixels included in the image.

In one embodiment, the color corrector 310 may extract a region corresponding to a fish using a predetermined fish region extraction algorithm. Here, the fish area extraction algorithm is learned to extract the area corresponding to the fish through a separate artificial neural network, or is a command programmed to extract the area corresponding to the fish by obtaining the boundary line between the fish and the background in the image. It may include.

In another embodiment, the color correction unit 310 may replace pixels with color values within a preset color range with preset color values so as not to affect the process of correcting the color of fish bones. For example, the color corrector 310 may replace pixels with a color value corresponding to the background color in the image with a pre-specified color value.

The image matching unit 320 can match multiple images taken in different environments for one fish into one image. At this time, the plurality of images may include a first image captured in a first lighting environment and a second image captured in a second lighting environment with lower brightness than the first lighting environment.

In one embodiment, the first image may be an image captured by setting a longer exposure time of the camera than the second image. In another embodiment, the first image may be an image captured by setting the sensitivity of the camera to be higher than that of the second image.

Accordingly, the color of the cross section of the fish in the first image may be clearer than that of the second image, and the borderline between meat and fat in the cross section of the fish may be clearer in the second image than in the first image.

Accordingly, when multiple images are input, the image registration unit 320 can match them by assigning different weights to each of the multiple images.

In one embodiment, the image registration unit 320 may register multiple images using an image blending technique. Here, the image blending technique may extract and match specific areas within each image in the process of matching multiple images.

To this end, the image matching unit 320 can match multiple images through Equations 1 and 2 below.

In Equation 1, g(x) may be the registered image, f0(x) may be the first image, and f1(x) may be the second image. Additionally, a may be a weight.

Therefore, referring to Equation 1, as the weight value of the matched image changes from 0 to 1, the influence of the first image may decrease and the influence of the second image may increase.

In Equation 2, dst(I) may be a registered image, src1(I) may be the first image, and src2(I) may be the second image. Additionally, alpha is (1-a) in Equation 1, beta is a, and gamma may be a preset weight.

Based on this, the image matching unit 320 adjusts the first image and the second image so that the cross-sectional color of the fish is emphasized in the first image among the plurality of images, and the boundary line between the meat and fat of the fish is emphasized in the second image. Weights can be assigned to .

In other words, the image registration unit 320 may obtain a registered image by combining the cross-sectional area of the first image and the border area of the second image.

In one embodiment, the cross-sectional color of the first image appears to have higher saturation compared to the border area of the second image, so the image registration unit 320 assigns a high weight to the saturation of the first image and the color of the second image. Low weight can be given to saturation.

In another embodiment, the cross-sectional color of the first image appears to have lower brightness compared to the border area of the second image, so the image registration unit 320 assigns a low weight to the brightness of the first image and High weight can be given to the brightness of .

In this regard, in one embodiment, the fish grade determination program 300 performs preprocessing by the color correction unit 310 when detecting the presence of fish bones in the image input through the input/output module 210. , If it is detected that multiple images of one fish have been input, preprocessing by the image matching unit 320 can be performed.

The rating determination unit 330 may input the preprocessed image into the fish rating neural network 400 to determine the rating of the fish.

At this time, the rating unit 330 may use one of the tail cut surface rating model and the midbone cut surface rating model included in the fish rating neural network 400 based on the user's selection.

Figures 4 and 5 are block diagrams showing an example of a process for determining the grade of fish in the grade determination unit of Figure 3.

Referring to FIG. 4, the tail cut surface rating model 231a is learned using the learning tail cut surface image 13a and the correct answer grade 15a through the artificial neural network learning device 100, and the fish rating device 200 ), it can be confirmed that the fish grade 50a is output by inputting the tail cut surface image 30a into the pre-learned tail cut surface grade determination model 231a.

Referring to FIG. 5, through the artificial neural network learning device 100, the midbone cut surface rating model (231b) is learned using the learning midbone cut surface image (13b) and the correct answer grade (15b), and the fish rating device (200) ), it can be confirmed that the grade 50b of the fish is output by inputting the midbone cut surface image 30b into the pre-learned midbone cut surface grade determination model 231b.

Figure 6 is a flowchart of an artificial neural network learning method according to an embodiment of the present invention.

Referring to FIGS. 1 and 6 , the processor 120 may receive a training image for a fish and a correct answer grade as label data for the training image through the input/output module 110 (S100).

Accordingly, the processor 120 may train the fish rating neural network to output a learning rating according to the learning image using the learning image and the correct answer rating (S200).

Figure 7 is a flowchart of a fish grading method according to an embodiment of the present invention.

Referring to FIGS. 2 and 7 , when an image of a cross-section of a fish is input through the input/output module 210, the processor 220 may preprocess the image (S600).

Accordingly, the processor 220 may determine the grade of the fish by inputting the pre-processed image into the pre-trained fish grade determination neural network 400 (S700).

Figures 8 and 9 are detailed flowcharts of preprocessing the image of Figure 7.

Referring to FIGS. 3 and 8 , the classification unit 311 may classify multiple pixels into multiple color groups (S610).

At this time, the processing unit 312 may check the first color group corresponding to the fish bone among the plurality of color groups (S620) and the second color group with the largest number of pixels among the plurality of color groups (S630). ).

Accordingly, the processing unit 312 may replace the color value of each of at least one pixel included in the first color group using the color value of at least one pixel included in the second color group (S640).

Referring to FIGS. 3 and 9, the image matching unit 320 determines a weight to emphasize the cross-sectional color of the fish in the first image (S660), and sets the weight to emphasize the border between the meat and fat of the fish in the second image. can be determined (S670).

Accordingly, the image matching unit 320 can match the first image and the second image by assigning different weights to each (S680).

Various embodiments of the present document are software (e.g., instructions stored in a machine-readable storage media (e.g., memory (built-in memory or external memory)) that can be read by a machine (e.g., a computer). : program). The device is a device capable of calling instructions stored in a storage medium and operating according to the called instructions, and may include an electronic device (eg, device) according to the disclosed embodiments. When the instruction is executed by a processor (eg, a processor), the processor may perform the function corresponding to the instruction directly or using other components under the control of the processor. Instructions may contain code generated or executed by a compiler or interpreter. A storage medium that can be read by a device may be provided in the form of a non-transitory storage medium. Here, 'non-transitory' only means that the storage medium does not contain signals and is tangible, and does not distinguish whether the data is stored semi-permanently or temporarily in the storage medium.

According to one embodiment, methods according to various embodiments disclosed in this document may be provided and included in a computer program product.

The above description is merely an illustrative explanation of the technical idea of the present invention, and those skilled in the art will be able to make various modifications and variations without departing from the essential quality of the present invention. Accordingly, the embodiments disclosed in the present invention are not intended to limit the technical idea of the present invention, but are for illustrative purposes, and the scope of the technical idea of the present invention is not limited by these embodiments. The scope of protection of the present invention shall be interpreted in accordance with the claims below, and all technical ideas within the scope equivalent thereto shall be construed as being included in the scope of rights of the present invention.

10: Training data
13: Image for training
13a: Tail cross-section image for training purposes
13b: Midbone cross section image for learning purposes
15: Correct Answer Grade
30: image
30a: tail cut surface image
30b: Midbone cross section image
50: Grade
100: Artificial neural network learning device
200: Fish grading device
231a: Tail cut surface grading model
231b: midbone cut surface grading model

Claims (18)

When an image of a cross-section of a fish is input, preprocessing the image; and
A fish grading method including; inputting the pre-processed image into a pre-trained fish grading neural network to determine the grading of the fish. The method of claim 1, wherein the fish grading neural network is:
It was learned according to the artificial neural network learning method,
The artificial neural network learning method is,
Inputting a learning image for a fish and a correct answer grade as label data for the learning image; and
A fish rating determination method including; using the learning image and the correct answer rating to train the fish rating neural network to output a learning rating according to the learning image. The method of claim 1, wherein the image is:
A fish grade determination method, which is one of a tail cut surface image obtained by photographing the tail cut surface of the fish and a midbone cut surface image obtained by photographing the midbone cut surface of the fish. The method of claim 1, wherein the pretreatment step includes:
Comprising: correcting the color value of a pixel corresponding to the bone of the fish among a plurality of pixels included in the image to the color value of a pixel corresponding to the flesh of the fish. The method of claim 4, wherein the step of correcting with the color value of the pixel corresponding to the flesh of the fish,
classifying the plurality of pixels into a plurality of color groups;
Identifying a first color group corresponding to the fish bone among the plurality of color groups;
identifying a second color group with the largest number of pixels among the plurality of color groups; and
Substituting the color value of each of at least one pixel included in the first color group using the color value of at least one pixel included in the second color group. A fish rating method comprising a. The method of claim 1, wherein the pretreatment step includes:
When multiple images are input,
A fish rating method including; assigning different weights to each of the plurality of images and matching them. The method of claim 6, wherein the step of matching by assigning different weights comprises:
Weighting is given to each of the first image and the second image so that the cross-sectional color of the fish is emphasized in the first image among the plurality of images, and the boundary line between the meat and fat of the fish is emphasized in the second image. How to grade fish. The method of claim 6, wherein the plurality of images are:
A fish rating method comprising a first image taken in a first lighting environment and a second image taken in a second lighting environment with lower brightness than the first lighting environment. Inputting a learning image for a fish and a correct answer grade as label data for the learning image; and
An artificial neural network learning method comprising: using the learning image and the correct answer grade to train a fish grade determination neural network to output a learning grade according to the learning image. The method of claim 9, wherein the learning image is:
An artificial neural network learning method, which is one of a learning tail cut surface image obtained by photographing the tail cut surface of the fish and a learning midbone cut surface image obtained by photographing the midbone cut surface of the fish. The method of claim 9, wherein the step of training the fish rating neural network is:
An artificial neural network learning method comprising: correcting the color value of a pixel corresponding to the bone of the fish among a plurality of pixels included in the learning image to the color value of a pixel corresponding to the flesh of the fish. The method of claim 9, wherein the step of training the fish rating neural network is:
When multiple training images are input,
An artificial neural network learning method comprising: assigning different weights to each of the plurality of training images and matching them. A memory in which a pre-trained fish grading neural network is stored; and
When an image of a cross-section of a fish is input, a processor that pre-processes the image and inputs the pre-processed image into the fish grade determination neural network to determine the grade of the fish. Fish grade determination device comprising a. A computer-readable recording medium storing a computer program,
When the computer program is executed by a processor,
When an image of a cross-section of a fish is input, preprocessing the image; and
A computer-readable recording medium comprising instructions for causing the processor to perform a method including inputting the pre-processed image into a pre-trained fish grade determination neural network to determine the grade of the fish. A computer program stored on a computer-readable recording medium,
When the computer program is executed by a processor,
When an image of a cross-section of a fish is input, preprocessing the image; and
A computer program comprising instructions for causing the processor to perform a method including inputting the pre-processed image into a pre-trained fish grade determination neural network to determine the grade of the fish. A memory storing learning images of fish and correct answer ratings as label data for the learning images; and
An artificial neural network learning device comprising; a processor that trains a fish rating neural network to output a learning rating according to the learning image using the learning image and the correct answer rating. A computer-readable recording medium storing a computer program,
When the computer program is executed by a processor,
Inputting a learning image for a fish and a correct answer grade as label data for the learning image; and
A computer-readable record comprising instructions for causing the processor to perform a method including: training a fish rating neural network to output a learning rating according to the learning image using the learning image and the correct answer rating. media. A computer program stored on a computer-readable recording medium,
When the computer program is executed by a processor,
Inputting a learning image for a fish and a correct answer grade as label data for the learning image; and
A computer program comprising instructions for causing the processor to perform a method including: training a fish rating neural network to output a learning rating according to the learning image using the learning image and the correct answer rating.

Images