JP2023054913A - Nishikigoi carp evaluation apparatus, method and program - Google Patents

Abstract

To provide a Nishikigoi carp evaluation apparatus, method, and program for objectively evaluating a Nishikigoi carp.SOLUTION: The Nishikigoi carp evaluation apparatus is provided with a feature quantity acquisition unit 11 for acquiring a feature quantity extracted from a photographed image of an evaluation target Nishikigoi carp; an evaluation unit 12 for evaluating the evaluation target Nishikigoi carp by inputting the feature quantity of the evaluation target Nishikigoi carp acquired by the feature quantity acquisition unit 11 to a learned model acquired by machine learning a relationship between the feature amounts of a plurality of the evaluation target Nishikigoi carps and an evaluation result of each Nishikigoi carp; and an output unit 13 for outputting the evaluation result of the evaluation target Nishikigoi carp evaluated by the evaluation unit 12.SELECTED DRAWING: Figure 1

Description

The present invention relates to a Nishikigoi evaluation apparatus, method, and program for evaluating the value of Nishikigoi.

Traditionally, competitions are held to evaluate Nishikigoi. At traditional Nishikigoi competitions, the Nishikigoi exhibited by the exhibitors are examined by judges and ranked based on the examination results.

Conventionally, Nishikigoi were judged from the following viewpoints.
1. Because Nishikigoi are living creatures, they must be "healthy."
3, Because it is also a work of art that is said to be a jewel that swims with ornamental fish, it must have artistry and breed characteristics.

Japanese Patent No. 6650984

However, conventional examinations are carried out based on vague criteria such as those mentioned above, so there is a high degree of dependence on the examiner's subjectivity. However, the specific content of the evaluation was not clear.

In the trade of Nishikigoi, the transaction amount is sometimes determined by the results of competitions, etc., and objective evaluation of Nishikigoi is desirable.

Incidentally, Patent Document 1 proposes a method for individual identification of Nishikigoi based on a photographed image of Nishikigoi, but does not propose anything about evaluation of Nishikigoi.

In view of the above circumstances, it is an object of the present invention to provide a Nishikigoi evaluation apparatus, method, and program capable of objectively evaluating Nishikigoi.

The Nishikigoi evaluation apparatus of the present invention includes a feature acquisition unit that acquires feature amounts extracted from an image of a Nishikigoi to be evaluated, and machine learning of the relationship between the feature amounts of a plurality of Nishikigoi and the evaluation results of each Nishikigoi. an evaluation unit for evaluating the Nishikigoi to be evaluated by inputting the feature values of the Nishikigoi to be evaluated obtained by the feature value obtaining unit to the trained model obtained by the above; and an output unit for outputting the evaluation result of the target Nishikigoi.

In the Nishikigoi evaluation method of the present invention, feature values extracted from images of Nishikigoi to be evaluated are acquired, and the relationship between the feature values of a plurality of Nishikigoi and the evaluation results of each Nishikigoi is obtained by machine learning. The Nishikigoi to be evaluated is evaluated by inputting the acquired feature values of the Nishikigoi to be evaluated to the trained model, and the evaluation result of the Nishikigoi to be evaluated is output.

The Nishikigoi evaluation program of the present invention comprises a step of acquiring feature values extracted from a photographed image of a Nishikigoi to be evaluated, and obtaining the relationship between the feature values of a plurality of Nishikigoi and the evaluation results of each Nishikigoi through machine learning. A step of evaluating the Nishikigoi to be evaluated by inputting the acquired feature values of the Nishikigoi to be evaluated and a step of outputting the evaluation result of the Nishikigoi to be evaluated to the trained model obtained by the computer. .

According to the Nishikigoi evaluation apparatus, method, and program of the present invention, the feature amount extracted from the photographed image of the Nishikigoi to be evaluated is acquired, and the relationship between the feature amount of a plurality of Nishikigoi and the evaluation result of each Nishikigoi is calculated mechanically. The Nishikigoi to be evaluated is evaluated by inputting the feature values of the Nishikigoi to be evaluated to the learned model obtained by learning, and the evaluation result of the Nishikigoi to be evaluated is output. Able to make objective evaluations.

A block diagram showing a schematic configuration of an embodiment of the Nishikigoi evaluation apparatus of the present invention. Diagram showing an example of an image of Nishikigoi to be evaluated A diagram showing an example of an image after posture correction processing A diagram showing an example of outline information of Nishikigoi A diagram showing an example of the color frequency distribution of each pixel that constitutes an image of Nishikigoi A diagram showing an example of a radar chart that is the evaluation result of Nishikigoi Flowchart for explaining the flow of processing by the Nishikigoi evaluation device

An embodiment of the Nishikigoi evaluation apparatus of the present invention will be described in detail below with reference to the drawings. FIG. 1 is a schematic configuration diagram of a Nishikigoi evaluation apparatus 1 of this embodiment.

A Nishikigoi evaluation apparatus 1 extracts a feature amount from an image of a Nishikigoi, evaluates the Nishikigoi based on the feature amount, and outputs the evaluation result. As a result, the apparatus can objectively evaluate Nishikigoi, which has conventionally been subjective.

Specifically, the Nishikigoi evaluation apparatus 1 includes an image acquisition unit 10, a feature amount acquisition unit 11, an evaluation unit 12, and an output unit 13, as shown in FIG.

The image acquisition unit 10 acquires an image of a Nishikigoi to be evaluated. Specifically, the image acquiring unit 10 acquires an image of the Nishikigoi viewed from above (dorsal fin side). In this embodiment, the dorsal fin side of Nishikigoi is the top, the ventral side is the bottom, the head side is the front, the tail side is the back, the right eye side is the right side, and the left eye side is the left side.

The image acquisition unit 10 acquires an image of Nishikigoi output from a terminal device connected to the Nishikigoi evaluation apparatus 1 via a communication line such as the Internet. The terminal device may be a tablet terminal, a smart phone, a desktop computer, a notebook computer, or the like.

The feature amount acquisition unit 11 extracts and acquires feature amounts from the image of the Nishikigoi to be evaluated acquired by the image acquisition unit 10 . In the present embodiment, information on the body shape of the Nishikigoi, information on the color of the Nishikigoi, information on the pattern of the Nishikigoi, and information on the finishing of the Nishikigoi are acquired as feature amounts.

Before performing the feature amount extraction process described above, the feature amount acquisition unit 11 performs preprocessing for increasing the accuracy of feature amount extraction on an image of the Nishikigoi to be evaluated. As preprocessing, the feature amount acquisition unit 11 first performs posture correction processing. Posture correction processing is processing for correcting the posture of Nishikigoi in the image so that it becomes a preset posture.

If the image of the Nishikigoi to be evaluated is, for example, an image having orthogonal X and Y axes as a coordinate system, as shown in FIG. This is a process to follow the Y direction when there is no such line. Specifically, in the posture correction process, first, a minimum rectangle (broken-line rectangle shown in FIG. 2) that includes the entire Nishikigoi contained in the image is set, and an image within that rectangle is cut out. Then, rotation processing is performed so that the long sides of the image within the rectangular range are parallel to the Y direction. By such clipping processing and rotation processing, an image after posture correction processing as shown in FIG. 3 is generated.

Next, color correction processing is applied to the image that has undergone attitude correction processing. This color correction process is a process for bringing the colors of an image closer to the primary colors. For example, if the image has a shadow, color correction processing is performed to remove the shadow. Existing image processing can be used for color correction processing for removing shadows.

Further, contour extraction processing is performed on the image that has undergone color correction processing. The outline extraction process is a process for extracting the outline of Nishikigoi. Existing image processing can also be used for this contour extraction processing. The outline information of the Nishikigoi extracted by the outline extraction process is used when extracting the feature quantity described below.

Based on the image subjected to the posture correction processing and the color correction processing as described above and the contour information of the Nishikigoi extracted by the contour extraction processing, the feature amount acquisition unit 11 obtains the body shape information, color information, and pattern information of the Nishikigoi. to obtain information on the product and information on the finish.

First, information on the body shape of Nishikigoi will be explained.

When evaluating Nishikigoi, the following points are taken into consideration regarding the body shape of Nishikigoi.
・The breech should not be too thin or too thick, and should be balanced in proportion to each part of the body. ・The size of the head and the whole body should not be unbalanced. Balanced and bilaterally symmetrical ・Arcuate with the dorsal muscles and backbone as the center line, bilaterally symmetrical and without bending ・The line from the head to the caudal fin should be smooth

Therefore, the feature quantity acquisition unit 11 of this embodiment obtains a to f shown in FIG. 4 based on the outline information of Nishikigoi. a is the width of the head, the distance from the inside of the left eye to the inside of the right eye. Also, b is the shoulder width (the width of the widest part of the body), which is the distance between the posterior end of the left pectoral fin and the posterior end of the right pectoral fin. c is the extent of protrusion of the belly, which is the distance between the left and right ends of the front end of the dorsal fin. d is the thickness of the breech and is the distance between the left and right ends of the front end of the caudal fin. e is the width of the pectoral fin, which is the distance from the root of the pectoral fin to the edge in the left-right direction. f is the vertical width of the pectoral fin, which is the length from the front end to the rear end of the pectoral fin.

The left and right eye positions, pectoral fin position, dorsal fin position and caudal fin position are detected by using existing image processing and pattern recognition.

Then, the feature quantity acquisition unit 11 calculates the following formulas to obtain the ratios Wa to Wf.
Wa=a/b, Wb=b/c, Wc=c/d, Wd=a/d, We=b/e,
Wf=e/f

Also, the feature quantity acquisition unit 11 obtains g through k shown in FIG. 4 based on the outline information of Nishikigoi. g is the total length of the body from the snout to the posterior end of the caudal fin. h is the length of the head from the tip of the mouth to the posterior end of the gill cover. i is the back length from the posterior end of the gill cover to the anterior end of the dorsal fin. j is the body length from the front edge of the dorsal fin to the front edge of the caudal fin. k is the length of the caudal fin from the anterior end of the caudal fin to the posterior end of the caudal fin.

Although not shown in FIG. 4, the position of the gill cover can be detected by extracting the outline of the gill. Further, when it is difficult to detect Nishikigoi only from an image taken from above, the image acquisition unit 10 further acquires an image taken from the left side or an image taken from the right side, and the feature amount acquisition unit 11 It is also possible to detect the position of the gill cover from the image.

Then, the feature quantity acquiring unit 11 calculates the following formulas to obtain the ratios Lh to Lf.
Lh=g/h, Li=g/i, Lj=g/j, Lk=g/k, Lf=g/f,

The feature quantity acquisition unit 11 calculates the ratios of the above-mentioned 11 items Wa to Wf and Lh to Lf as the body shape information of Nishikigoi.

Also, the feature quantity acquisition unit 11 obtains the line of the spine of the Nishikigoi from the outline information of the Nishikigoi. Specifically, the line from the front end of the dorsal fin through the dorsal fin to the front end of the caudal fin is determined as the spine line. The rear end of the dorsal fin to the front end of the caudal fin may be connected, for example, with a straight line.

Next, the feature quantity acquisition unit 11 acquires information on the presence or absence of defects in the Nishikigoi as information on the body shape of the Nishikigoi.

Specifically, based on the outline information of Nishikigoi, the feature quantity acquiring unit 11 confirms whether or not a total of four mustaches, two large and two small, are present. Also, the feature quantity acquiring unit 11 obtains the positions of the left and right eyes from the Nishikigoi image, and confirms whether or not the positions of the eyes are bilaterally symmetrical. Also, the feature quantity acquiring unit 11 detects whether or not the gill covers are warped based on the outline information of the Nishikigoi. Also, the feature quantity acquisition unit 11 confirms whether or not the pectoral fins are missing, twisted, or deformed, based on the outline information of the Nishikigoi.

Also, the feature quantity acquisition unit 11 extracts scales from the Nishikigoi image and checks whether or not the arrangement of the scales is irregular. Specifically, the feature amount acquisition unit 11 obtains the center of each scale, and obtains a straight line connecting the centers of adjacent scales in the front-rear direction. Then, the feature amount acquisition unit 11 detects that the straight line is not smooth and includes irregularities, thereby confirming the unevenness of the scales.

In addition, the feature amount acquiring unit 11 confirms that the scale pattern extracted from the Nishikigoi image does not have a scale pattern within a range equal to or greater than a preset area, thereby obtaining information on the presence or absence of keloids. to get

As information on the presence or absence of defects in Nishikigoi, the feature quantity acquisition unit 11 uses mustache information, left and right eye symmetry information, gill cover warpage information, pectoral fin defect information, and scale information as described above. Acquire alignment information and keloid information.

That is, the feature quantity acquisition unit 11 acquires the above-mentioned 11 ratios, the spine line, and 6 pieces of defect information as the body shape information of Nishikigoi.

As information on the spine line, the coordinates of a plurality of points on the line when the spine line is plotted on a coordinate system set in advance, for example, are acquired. For the six pieces of defect information, for example, a numerical value of "1" is obtained as information indicating that there is a defect, and a numerical value of "0" is obtained as information indicating that there is no defect.

Next, the color information of Nishikigoi will be explained. In this embodiment, color information when the type of Nishikigoi is "Kohaku" will be described.

When evaluating Nishikigoi (Kohaku), the following points are taken into consideration regarding the color information of Nishikigoi, with regard to the white background (= background).
・Deep, thick milky white color with a sense of transparency

Therefore, the feature quantity acquisition unit 11 of the present embodiment obtains the color frequency distribution of each pixel forming the Nishikigoi image, as shown in FIG.

Then, the feature amount acquisition unit 11 obtains the frequency of milky-white pixels based on the color frequency distribution and acquires it as white background information. It is assumed that the RGB values of milky white are set in advance. The white background of Nishikigoi (Kohaku) is preferably milky white.

In addition, the feature amount acquisition unit 11 obtains the frequency of pink pixels corresponding to the lower scarlet color based on the color frequency distribution, and obtains the lower scarlet information. It is assumed that the RGB values of pink corresponding to Shimohi are set in advance. The shiroji of Nishikigoi (Kohaku) is preferably milky white, and it is preferable that pink is not visible.

Further, the feature amount acquisition unit 11 obtains the frequency of yellow pixels corresponding to yellowing based on the color frequency distribution, and acquires it as yellowing information. It is assumed that the RGB values of yellow corresponding to yellowing are set in advance. It is preferable that there is no yellowing.

In addition, the feature amount acquisition unit 11 obtains blemish information by obtaining the frequency of brownish and blackish pixels corresponding to the blemish based on the color frequency distribution. It is assumed that brownish and blackish RGB values corresponding to blemishes are set in advance. It is preferable that there is no stain.

Next, regarding information on the color of Nishikigoi, the following points are taken into account with regard to Hiban (red part).
・The color of the hiban should be uniform from the head to the tail. ・The red color should be clear and the color should be good. Seeing the thickness of the hiban (kneading, that is, the one with no visible scales is considered to be thick)
・The hiban must be well-cut and well-ordered, and the kiwa of the tohiban and the kiwa and marbling of the scales must not bleed.

Therefore, the feature amount acquisition unit 11 of the present embodiment extracts the hiban range from the Nishikigoi image, and calculates the color frequency distribution of each pixel in the hiban range. It is assumed that the range of RGB corresponding to the range of Hiban is set in advance.

Then, the feature quantity acquisition unit 11 acquires information on the uniformity of the color of the hiban by obtaining the dispersion or standard deviation of the frequency distribution of the color in the range of the hiban. A smaller dispersion or standard deviation is preferred because it provides uniformity of color.

Further, the feature amount acquisition unit 11 converts the RGB of each pixel in the range of scarlet board, for example, into a signal of the color space of L ^* a ^* b ^* , and obtains the saturation based on the values of a ^* and b ^* . Acquired as information on the sharpness (hue) of the color of Hiban.

In addition, the feature amount acquisition unit 11 extracts the scale pattern contained in the entire image of the hiban range, detects the edge of the extracted scale pattern, obtains the edge amount, and uses it as information on the thickness of the hiban. get. The edge amount is the total number of pixels forming an edge. It is preferable that the edge amount is small because the scarlet board has a thickness.

Further, based on the Nishikigoi image, the feature amount acquisition unit 11 obtains edge sharpness (degree of blurring) at the boundary between the range of the scarlet board and the range of the white background, and acquires it as information on the bleeding of the marbling and kiwa. . An existing method can be used to calculate the edge sharpness (degree of blurring). The higher the edge sharpness, the less bleeding of margins and wrinkles, which is preferable.

It should be noted that the fact that there are no blemishes or blemishes in the scarlet board is acquired as the aforementioned blemish information.

That is, the feature amount acquisition unit 11 obtains, as color information of Nishikigoi, information on white background, information on shimohi, information on yellowing, information on spots, information on uniformity of color of hiban, sharpness of color of hiban (hue) information, information on the thickness of the hiban, and information on the bleed of the sashimi edge.

Next, information on the pattern of Nishikigoi will be described. In the present embodiment, pattern information when the type of Nishikigoi is "Kohaku" will be described.

When evaluating Nishikigoi (Kohaku), the following points are taken into account with respect to information on Nishikigoi patterns.
・At each part from the head to the body to the tail, the shape, size and number of the spots should be in harmony with the body.

Therefore, the feature amount acquisition unit 11 of the present embodiment extracts the range of hiban from the Nishikigoi image.
Then, the feature amount acquisition unit 11 determines whether part of the range of Hiban is included in the range of the tip of the nose,
Check whether the range of the scarlet board is approaching the base of the tail. Note that the range of the tip of the nose and the base of the tail are set using an existing image processing method based on preset conditions.

If part of the range of Hiban is not included in the range of the tip of the nose and the range of Hiban does not reach the base of the tail, the feature amount acquisition unit 11 determines "good ” information, and if part of the range of Hiban is included in the range of the tip of the nose, or if the range of Hiban is approaching the base of the tail, “bad” as information on the hook of Hiban Get information about

In addition, the feature quantity acquisition unit 11 divides the Nishikigoi image into three parts, ie, the head, the body, and the tail, and specifies the range of hiban included in each part. The head, body and tail may be divided according to preset conditions.

Then, the feature amount acquiring unit 11 calculates the ratio of the area of the scarlet board included in each part to the total area of each part, and acquires the ratio of each part as balance information of the scarlet board. It should be noted that when the hiban ratio of each part is 50% to 90%, it is judged that the hiban balance is good, and in the case of other ratios, it is judged that the hiban balance is bad.

In addition, the feature amount acquisition unit 11 acquires the number of hiban included in each part and the circumference length of each hiban as information on the balance of the hiban.

In other words, the feature quantity acquisition unit 11 acquires the information about the pattern of the Nishikigoi, and the information about the balance of the hiban. As for the hiban kake information, for example, if the kakehan information is "good", a numerical value of "1" is obtained, and if it is "bad", a numerical value of "0" is obtained.

Next, information on finishing of Nishikigoi will be explained. In the present embodiment, the finish information when the type of Nishikigoi is "Kohaku" will be described.

When evaluating Nishikigoi (Kohaku), the following points are taken into consideration regarding the finished information of Nishikigoi.
・The color must be clear and vivid, with gloss and shine, and whether or not there is cloudiness. (should be light in color)

Therefore, the feature amount acquisition unit 11 obtains the contrast, brightness, and saturation based on the Nishikigoi image, and acquires them as finish information. As for the chroma, the chroma of the Hiban range is obtained as the color information of the Nishikigoi, but as the finish information, the chroma of the entire Nishikigoi image is obtained.

The above is the description of the feature quantity acquired by the feature quantity acquisition unit 11 of the present embodiment.

Next, the evaluation unit 12 will be described.

The evaluation unit 12 has a trained model obtained by machine-learning the relationship between the above-described feature values of a plurality of Nishikigoi and the evaluation results of each Nishikigoi. More specifically, the evaluation unit 12 of the present embodiment includes a body type evaluation trained model for evaluating the body type of the Nishikigoi to be evaluated, a quality evaluation trained model for evaluating the quality of the Nishikigoi, It has four trained models: a pattern evaluation trained model for evaluating the pattern of Nishikigoi and a finished evaluation trained model for evaluating the finish of Nishikigoi.

As for the trained model for body shape evaluation, for example, the above-mentioned body shapes of 10 Nishikigoi that became world champions (overall overall winners) at past International Nishikigoi Shows and All Japan Nishikigoi Shows sponsored by All Japan Airinkai and All Japan Shinkokai. Machine learning is performed on the relationship between the information and the evaluation results (scores) related to the body shape, and a learned model that serves as a reference is generated. Then, for a learned model that serves as a reference, machine learning is performed on the relationship between the above body shape information of a plurality of arbitrary Nishikigoi and the evaluation results (scores) related to the body shape to obtain a learned model for body shape evaluation. The arbitrary plurality of Nishikigoi described above include Nishikigoi with various evaluation results, such as those with good evaluation results and those with poor evaluation results.

As for the quality evaluation trained model, machine learning is performed on the relationship between the above-described color information of the 10 world champion Nishikigoi and the evaluation results (scores) related to the color to generate a reference trained model. Then, for the reference trained model, machine learning is performed on the relationship between the aforementioned color information of a plurality of arbitrary Nishikigoi and the evaluation results (scores) related to the color to obtain a trained model for quality evaluation.

As for the pattern evaluation trained model, machine learning is performed on the relationship between the above-described pattern information of the ten world champion Nishikigoi and the evaluation results (scores) of the pattern to generate a reference trained model. Then, for a learned model that serves as a reference, a learned model for pattern evaluation is obtained by machine-learning the relationship between the above-described pattern information of a plurality of arbitrary Nishikigoi and the evaluation results (scores) related to the pattern.

The learned model for finish evaluation performs machine learning on the relationship between the above-described information on the finish of the 10 world champion Nishikigoi and the evaluation results (scores) related to the finish, and generates a learned model that serves as a reference. Further, for a learned model serving as a reference, a learned model for finish evaluation is obtained by machine-learning the relationship between the finish information of a plurality of arbitrary Nishikigoi and the evaluation results (scores) related to the finish.

As a machine learning method, a known method can be used, such as a deep neural network (DNN), a convolutional neural network (CNN), a recurrent neural network (RNN), and a denoising stack autoencoder (DSA). can be used.

The evaluation unit 12 evaluates the Nishikigoi to be evaluated using four trained models: a trained model for body type evaluation, a trained model for quality evaluation, a trained model for pattern evaluation, and a trained model for finish evaluation. .

Specifically, the evaluation unit 12 inputs the body shape information of the Nishikigoi to be evaluated acquired by the feature value acquisition unit 11 to the trained model for body shape evaluation, thereby obtaining the body shape of the Nishikigoi to be evaluated. Evaluate and obtain evaluation results (scores).

In addition, the evaluation unit 12 evaluates the quality of the Nishikigoi to be evaluated by inputting the color information of the Nishikigoi to be evaluated acquired by the feature value acquisition unit 11 to the trained model for quality evaluation, Get the evaluation result (score).

In addition, the evaluation unit 12 evaluates the pattern of the Nishikigoi to be evaluated by inputting information of the pattern of the Nishikigoi to be evaluated acquired by the feature value acquisition unit 11 to the trained model for pattern evaluation, Get the evaluation result (score).

In addition, the evaluation unit 12 evaluates the finish of the Nishikigoi to be evaluated by inputting information on the finish of the Nishikigoi to be evaluated acquired by the feature amount acquisition unit 11 to the learned model for finish evaluation, Get the evaluation result (score).

As described above, the evaluation unit 12 acquires the body shape score, color score, pattern score, and finish score of the Nishikigoi to be evaluated.

Furthermore, the evaluation unit 12 multiplies and adds preset weights to the above-described four scores to obtain a comprehensive evaluation result (total score) of the Nishikigoi to be evaluated.

The weighting can be arbitrarily set, but for example, 40% for the body shape score, 30% for the quality score, 20% for the pattern score, and 10% for the finish score. be done. In this case, if the figure score is s1, the quality score is s2, the pattern score is s3, and the finishing score is s4, the total score sa is calculated by the following formula.
sa = 0.4 x s1 + 0.3 x s2 + 0.2 x s3 + 0.1 x s4

Note that the weighting described above is not limited to the above example, and the weighting may be changed according to, for example, the growth stage of the Nishikigoi to be evaluated.

For example, if the Nishikigoi to be evaluated is at the juvenile stage ((part 12-25) 1cm-25cm), the weighting is set to 25% body type, 25% quality, 25% pattern, and 25% finish. . In addition, when the Nishikigoi to be evaluated is at the stage of Young Carp ((Part 30 to 40) 25.1 cm to 40 cm), the weighting is 35% for body type, 30% for quality, 20% for pattern, and 15% for finish. is set to In addition, when the Nishikigoi to be evaluated is at the stage of adult fish ((Part 45 to 55) 40.1 cm to 55 cm), the weighting is 40% for body shape, 30% for quality, 20% for pattern, and 10% for finish. set. In addition, when the Nishikigoi to be evaluated is at the stage of fish ((Part 60 to 70) 55 cm to 70 cm), the weighting is 40% for body type, 30% for quality, 20% for pattern, and 10% for finish. is set to In addition, when the Nishikigoi to be evaluated is at the stage of Kyogoi ((75th to 90th division) 70.1cm to 90cm), the weighting is 50% for body type, 25% for quality, 15% for pattern, and 10% for finish. set.

The evaluation unit 12 may receive information on the growth stages of Nishikigoi described above, set weights according to the stages, and calculate the above-described total score sa.

In this embodiment, the evaluation unit 12 has four trained models, but the four trained models are connected to the Nishikigoi evaluation device 1 via a communication network. It may be stored in a server device or the like, which is used for evaluation.

In addition, the Nishikigoi evaluation apparatus 1 of the present embodiment is provided with a trained model generation unit, and in the trained model generation unit, a trained model for body type evaluation, a trained model for quality evaluation, a trained model for pattern evaluation, and a finished model are generated. It is also possible to generate four trained models for the evaluation trained model. In that case, the image acquisition unit 10 acquires the above-described Nishikigoi image for machine learning, and the feature amount acquisition unit 11 acquires the feature amount of the acquired image. Moreover, the evaluation results (scores) of each Nishikigoi may be set and input by the user using a predetermined input device (not shown) connected to the Nishikigoi evaluation apparatus 1 .

The output unit 13 outputs the evaluation result of the evaluation target Nishikigoi evaluated by the evaluation unit 12 . The output unit 13 of the present embodiment outputs the body shape score s1, the quality score s2, the pattern score s3, the finish score s4, and the overall score sa of the Nishikigoi to be evaluated.

Specifically, the output unit 13 of this embodiment outputs the body shape score s1, quality score s2, pattern score s3, finish score s4, and overall score sa to a predetermined display device (not shown). ). As a display method, for example, as shown in FIG. 6, the figure score s1, the quality score s2, the pattern score s3, and the finish score s4 are displayed as a radar chart, and the overall score sa is displayed as a text.

The output destination of the output unit 13 of this embodiment is not limited to the display device, and may be output to a computer, server device, or the like connected to the Nishikigoi evaluation device 1 via a communication network. , and may be output to a printing device such as a printer.

The Nishikigoi evaluation apparatus 1 is composed of a computer, and includes a CPU (Central Processing Unit) or GPU (Graphics Processing Unit), semiconductor memories such as ROM (Read Only Memory) and RAM (Random Access Memory), storage such as a hard disk, and communication. It has an I/F (Interface) and the like.

A colored carp evaluation program according to an embodiment of the present invention is installed in the semiconductor memory or hard disk of the colored carp evaluation device 1 . When the CPU or GPU executes this Nishikigoi evaluation program, the above-described image acquisition unit 10, feature amount acquisition unit 11, evaluation unit 12, and output unit 13 function.

That is, the Nishikigoi evaluation program is obtained by performing machine learning of the step of obtaining the feature values extracted from the photographed image of the Nishikigoi to be evaluated, and the relationship between the feature values of a plurality of Nishikigoi and the evaluation results of each of the Nishikigoi. A computer is caused to execute a step of evaluating the Nishikigoi to be evaluated by inputting the characteristic amount of the Nishikigoi to be evaluated and a step of outputting the evaluation result of the Nishikigoi to be evaluated to the trained model.

Note that the hardware configuration of the Nishikigoi evaluation apparatus 1 is not limited to the configuration described above.

Further, in the present embodiment, all the functions of the image acquisition unit 10, the feature amount acquisition unit 11, the evaluation unit 12, and the output unit 13 are executed by the Nishikigoi evaluation program. Alternatively, all functions may be configured from hardware such as ASIC (Application Specific Integrated Circuit), FPGA (Field-Programmable Gate Array), and other electric circuits.

Next, the processing flow of the Nishikigoi evaluation apparatus 1 of this embodiment will be described with reference to the flowchart shown in FIG.

First, an image of a Nishikigoi to be evaluated is acquired by the image acquisition unit 10 (S10).

Next, in the feature quantity acquisition unit 11, the above-described pre-processing is performed on the image of Nishikigoi to be evaluated (S12).

Subsequently, the feature amount acquisition unit 11 acquires the feature amounts of body shape information, color information, pattern information, and finishing information of Nishikigoi based on the preprocessed image (S14).

Then, the feature values for each of the four trained models possessed by the evaluation unit 12 are input, and the body shape score, quality score, pattern score, finish score, and total score of the Nishikigoi to be evaluated are obtained ( S16).

The score of the Nishikigoi to be evaluated obtained by the evaluation unit 12 is output to the output unit 13, and the output unit 13 outputs the score of the body shape, the quality, the pattern, and the finish of the Nishikigoi to be evaluated as a radar chart. display, and display the total score as text. (S18).

In addition, in the Nishikigoi evaluation apparatus 1 of the above-described embodiment, four trained models are used: a trained model for body shape evaluation, a trained model for quality evaluation, a trained model for pattern evaluation, and a trained model for finish evaluation. Nishikigoi are evaluated, but not limited to this, the relationship between the body shape information, color information, and pattern information of multiple Nishikigoi and the overall score of each Nishikigoi is machine-learned and learned for comprehensive evaluation. A model may be generated. Then, the evaluation unit 12 may obtain a total score by inputting body shape information, color information, and pattern information of the Nishikigoi to be evaluated to the learned model for comprehensive evaluation. In addition, a learned model for comprehensive evaluation is generated by adding finishing information as a feature amount, and by inputting body shape information, color information, pattern information, and finishing information of the Nishikigoi to be evaluated, the overall score is calculated. may be requested.

Also, in the description of the above embodiment, the evaluation method for the case where the type of Nishikigoi is "Kohaku" has been described, but the Nishikigoi evaluation apparatus 1 may evaluate other types of Nishikigoi. That is, the Nishikigoi evaluation apparatus 1 may evaluate a plurality of types of Nishikigoi. In this case, the feature quantity acquisition unit 11 of the Nishikigoi evaluation apparatus 1 acquires a feature quantity corresponding to the type of Nishikigoi, and the evaluation unit 12 evaluates the Nishikigoi using a trained model corresponding to the type of Nishikigoi. That is, the evaluation unit 12 has a plurality of trained models corresponding to a plurality of types of Nishikigoi. As with the above-described embodiment, the trained models corresponding to the types of Nishikigoi are a trained model for body shape evaluation, a trained model for quality evaluation, a trained model for pattern evaluation, and a finished evaluation for each type of Nishikigoi. It has 4 trained models for training models.

As for the types of Nishikigoi, the above-mentioned 1. Red and white, 2. Taisho Sanshoku, 3. Showa Sanshoku, 4. 5. photogenic things; 5. Gold and silver scale type 1 (1 to 4 gold and silver scale types); 7. tortoiseshell; Asagi, 8. 9. Shusui; 10. clothes; oddities, 11. Goshiki, 12. 13. Hikari-moyo-mono (light-patterned in the Kokuryu system); Hikari Utsurimono, 14. 15. Two types of gold and silver scales (Kinginrin 2shu) (6 to 13 types of gold and silver scales); 16. plain ones (including silver scale and pine needles); 17. Hikari Mujimono (including Ginrin and Matsuba); 18. Tancho (all varieties with a tancho pattern, including silver scales); 19. male red and white; Male Taisho tricolor, 20. male Showa Sanshoku, and 21 . There are 21 types of male gold and silver scales (18 to 21 types of gold and silver scales).

As mentioned above, the feature values differ depending on the type of Nishikigoi. For example, in the case of Taisho Sanshoku, ink is included. is added. As the black uniformity information, for example, the variance and standard deviation of the pixels of the color corresponding to black are obtained. Also, as information on the termination of black, for example, the dispersion of the positions of independent black patterns corresponding to black is obtained, and whether or not the patterns are scattered is evaluated. As information on the thickness of the black, an edge amount within the range of the black is obtained.

In addition, in the case of Showa Sanshoku, as information on the color of Nishikigoi, information on whether or not the base of the sunken ink is Asagiji is added. As the information as to whether the background of the ink is pale yellow, a pixel of a color corresponding to the range of the ink is obtained.

In this way, a unique feature amount is acquired for each type of Nishikigoi, a trained model is generated, and evaluation is performed according to the type of Nishikigoi.

Regarding the Nishikigoi evaluation apparatus of the present invention, the following additional remarks are disclosed.
(Appendix)

In addition, in the Nishikigoi evaluation apparatus of the present invention, the feature amount acquisition unit can acquire body shape information of the Nishikigoi as the feature amount.

In addition, in the Nishikigoi evaluation apparatus of the present invention, the feature amount acquiring section can acquire color information of the Nishikigoi as the feature amount.

In addition, in the Nishikigoi evaluation apparatus of the present invention, the feature amount acquisition section can acquire information on the pattern of the Nishikigoi as the feature amount.

In addition, in the Nishikigoi evaluation apparatus of the present invention, the feature quantity acquisition unit can acquire body shape information, color information, and pattern information of Nishikigoi as feature quantities.

In addition, in the Nishikigoi evaluation apparatus of the present invention, the evaluation unit can have a trained model for each kind of Nishikigoi.

1 Nishikigoi Evaluation Apparatus 10 Image Acquisition Unit 11 Feature Amount Acquisition Unit 12 Evaluation Unit 13 Output Unit

Claims (8)

a feature quantity acquisition unit that acquires a feature quantity extracted from an image of a Nishikigoi to be evaluated;
The feature values of the Nishikigoi to be evaluated acquired by the feature value acquisition unit are obtained from a learned model obtained by machine learning the relationship between the feature values of a plurality of Nishikigoi and the evaluation results of each of the Nishikigoi. an evaluation unit that evaluates the Nishikigoi to be evaluated by inputting;
and an output unit for outputting evaluation results of the Nishikigoi to be evaluated evaluated by the evaluation unit. 2. The Nishikigoi evaluation apparatus according to claim 1, wherein said feature quantity acquisition unit acquires body shape information of Nishikigoi as said feature quantity. 3. The Nishikigoi evaluation apparatus according to claim 1 or 2, wherein said feature quantity acquiring unit acquires color information of Nishikigoi as said feature quantity. 4. The Nishikigoi evaluation apparatus according to any one of claims 1 to 3, wherein the feature quantity acquisition unit acquires pattern information of Nishikigoi as the feature quantity. 2. The Nishikigoi evaluation apparatus according to claim 1, wherein said feature quantity acquisition unit acquires body shape information, color information and pattern information of Nishikigoi as said feature quantity. The Nishikigoi evaluation apparatus according to any one of claims 1 to 5, wherein the evaluation unit has the learned model for each kind of Nishikigoi. Acquire the feature value extracted from the image of Nishikigoi to be evaluated,
The evaluation by inputting the acquired feature amount of the Nishikigoi to be evaluated to a learned model obtained by machine learning the relationship between the feature amount of a plurality of Nishikigoi and the evaluation result of each Nishikigoi. Evaluate the target Nishikigoi,
A Nishikigoi evaluation method for outputting an evaluation result of the Nishikigoi to be evaluated. a step of acquiring a feature amount extracted from an image of a Nishikigoi to be evaluated;
The evaluation by inputting the acquired feature amount of the Nishikigoi to be evaluated to a learned model obtained by machine learning the relationship between the feature amount of a plurality of Nishikigoi and the evaluation result of each Nishikigoi. evaluating a target Nishikigoi;
A Nishikigoi evaluation program that causes a computer to execute the step of outputting the evaluation result of the Nishikigoi to be evaluated.

Images