KR20240057753A - System for analyzing health status of pet using stool image - Google Patents

Abstract

The present invention relates to a companion animal health status analysis system using feces images. The companion animal health status analysis system using stool images according to one aspect of the present invention is obtained by photographing the stool of the companion animal, and is located in the location area of a preset object corresponding to at least one of normal stool, bloody stool, mucous stool, and hookworm. A database that stores learning images containing labeling data for; An image input unit that receives images obtained by photographing the feces of a companion animal from the user terminal; Training data including learning images stored in the database are applied to detect objects corresponding to at least one of normal stool, bloody stool, mucous stool, and hookworm in the image. When an object is detected, the detected object is classified into normal stool, bloody stool, mucous stool, and hookworm. A learning model that is pre-trained so that it can be classified into at least one of the following; an object detection and classification unit that detects an object from an image input to the image input unit using the learning model and, when an object is detected, classifies the object into at least one of normal stool, bloody stool, mucous stool, and hookworm using the learning model; and a notification providing unit that outputs a preset notification to the user terminal based on at least one of the object detection result and the object classification result of the object detection and classification unit.

Description

Pet health status analysis system using stool images {SYSTEM FOR ANALYZING HEALTH STATUS OF PET USING STOOL IMAGE}

The present invention relates to a companion animal health status analysis system using feces images that can determine whether there is an abnormality in the companion animal's health status using images obtained by photographing the companion animal's stool.

Recently, not only is the number of people raising companion animals increasing, but attachment to and interest in companion animals is increasing day by day, and accordingly, interest in the health of companion animals is also increasing.

When a pet has a health problem, it cannot communicate to the person who owns the pet. Therefore, in order to smoothly check the health of the pet, there is an inconvenience of having to visit a medical facility for an examination.

However, even if you visit a medical facility, the cost of examining your pet could be burdensome, and accordingly, if your pet develops a health problem and behaves abnormally, and the person raising the pet becomes aware of this, the pet's health checkup may be necessary. This happened frequently.

In addition, when a companion animal exhibits abnormal behavior due to a health problem, the disease is often found to have already progressed to an advanced stage, resulting in excessive treatment costs.

In order to solve the above-mentioned problems, devices and systems have been developed to determine and manage the health status of companion animals. However, devices and systems according to the prior art are equipped with a separate device to check the health status of companion animals. There was the inconvenience of having to do it.

KR 10-2330858 B1 KR 10-2021-0106773 A

The present invention was invented to solve the above problems, and is a stool image that can analyze the health status of a companion animal by analyzing the image obtained by photographing the companion animal's stool so that the health status of the companion animal can be easily checked. The purpose is to provide a companion animal health status analysis system using .

The object of the present invention is not limited to the object mentioned above, and other objects not mentioned can be clearly understood from the description below and can be fully included in the object of the present invention.

In one aspect of the present invention to achieve the above object, the companion animal health status analysis system using feces images is obtained by photographing the feces of the companion animal, and a preset object corresponding to at least one of normal stool, bloody stool, mucous stool, and hookworm. A database that stores learning images containing labeling data for the location area; An image input unit that receives images obtained by photographing the feces of a companion animal from the user terminal; By applying training data including learning images stored in the database, an object corresponding to at least one of normal stool, bloody stool, mucous stool, and hookworm is detected in the image, and when an object is detected, the detected object is classified into normal stool, bloody stool, mucous stool, and hookworm. A learning model that is pre-trained so that it can be classified into at least one of the following; an object detection and classification unit that detects an object from an image input to the image input unit using the learning model and, when an object is detected, classifies the object into at least one of normal stool, bloody stool, mucous stool, and hookworm using the learning model; and a notification providing unit that outputs a preset notification to the user terminal based on at least one of the object detection result and the object classification result of the object detection and classification unit.

The present invention with the above configuration can expect the following effects.

By receiving the image obtained by photographing the pet's feces and determining the state of the pet's feces by applying the input image to a pre-trained learning model, the health status of the pet can be determined. , it has the effect of allowing you to simply analyze the health status of your pet without the need for a separate device.

In particular, learning images can be preprocessed so that the learning model can effectively detect objects corresponding to hookworms present in companion animal feces. As the learning model is pre-trained using the preprocessed learning images, the learning model is input. Objects corresponding to hookworm can be effectively detected in the image, which can help users more quickly recognize health problems in pets due to hookworm outbreaks.

Figure 1 is a block diagram illustrating a companion animal health status analysis system using feces images according to an embodiment of the present invention.
Figure 2 is a flowchart for explaining the algorithm of the image extraction unit and the image preprocessing unit of the companion animal health status analysis system using feces images according to an embodiment of the present invention.

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 will be implemented in various different forms, but the present embodiments only serve to ensure that the disclosure of the present invention is complete, and those skilled in the art It is provided to fully inform the user of the scope of the invention. Meanwhile, the terms used in this specification are for describing embodiments and are not intended to limit the present invention. As used herein, singular forms also include plural forms, unless specifically stated otherwise in the context.

Hereinafter, the pet health status analysis system using feces images according to an embodiment of the present invention will be described in detail with reference to the attached drawings.

The companion animal health status analysis system 10 using feces images according to an embodiment of the present invention includes a database 100, an image input unit 200, a learning model 300, an object detection and classification unit 400, and a notification provision unit. It includes (500) and a communication unit (600).

The companion animal health status analysis system 10 using feces images according to an embodiment of the present invention may communicate with the user terminal 20 using a preset communication connection method. At this time, the communication connection method between the pet health status analysis system 10 using the feces image and the user terminal 20 may be communication using an external communication network.

The user terminal 20 includes an input unit 21 that receives input values according to the user's input, a screen unit 22 that can display preset text and an interface, and a sound output unit that can output preset sounds to the outside. (23) and a terminal communication unit 24 that communicates with the companion animal health status analysis system 10 using feces images using an external communication network.

The terminal communication unit 24 of the user terminal 20 may be connected to the communication unit 600 through an external communication network.

The user terminal 20 is not limited as long as it is a terminal equipped with the above-described input unit 21, screen unit 22, sound output unit 23, and terminal communication unit 24. For example, a smartphone, tablet PC, It may be any one of a PDA, a personal computer, and a laptop computer.

The user terminal 20 may receive an image according to the user's input and transmit it to the companion animal health status analysis system 10 using the feces image.

The user terminal 20 may have preset software that is linked to the companion animal health status analysis system 10.

The database 100 is obtained by photographing the stool of a companion animal, and stores learning images containing labeling data for the location area of a preset object corresponding to at least one of normal stool, bloody stool, mucous stool, and hookworm. You can.

The database 100 is obtained by photographing a pet dog's stool, and includes a location area of a preset object corresponding to at least one of normal stool, bloody stool, mucous stool, and hookworm, and labeling that includes an annotation corresponding to the location area. It may be a pre-stored learning image containing data. At this time, the annotation corresponding to the labeling data may correspond to at least one of normal stool, bloody stool, mucous stool, and hookworm, and may preferably correspond to any one of normal stool, bloody stool, mucous stool, and hookworm.

For example, if the learning image pre-stored in the database 100 includes labeling data including a location area of a preset object corresponding to hookworm and an annotation corresponding to the location area, the annotation included in the labeling data is related to hookworm. It may be a correspondence.

As for the learning image pre-stored in the database 100, the learning model 300, which will be described later, detects an object corresponding to at least one of normal stool, bloody stool, mucous stool, and hookworm in the image, and classifies the detected object into normal stool, bloody stool, mucus stool, and It may be included in training data to classify it as at least one of hookworms.

The learning image previously stored in the database 100 may be included in training data to enable the learning model 300, which will be described later, to detect an object corresponding to hookworm in the image.

The image input unit 200 may receive an image according to a user's input.

The image input unit 200 may receive an image from the user terminal 20 according to the user's input into the user terminal 20.

The learning model 300 applies training data including learning images previously stored in the database 100 to detect an object corresponding to at least one of normal stool, bloody stool, mucus stool, and hookworm in the image, and when the object is detected, the detected object It may be previously learned so that it can be classified into at least one of normal stool, bloody stool, mucous stool, and hookworm.

The learning model 300 uses a preset deep learning model to detect an object corresponding to at least one of normal stool, bloody stool, mucous stool, and hookworm. When an object is detected, the detected object is classified into one of normal stool, bloody stool, mucous stool, and hookworm. It may be pre-learned to be able to classify at least one thing, and in this case, the pre-set deep learning model is generally not limited as long as it can detect and classify objects in the image, for example, YOLO, v3, SSD, and Faster RCCN. It can be any one of them.

The object detection and classification unit 400 detects an object corresponding to at least one of normal stool, bloody stool, mucous stool, and hookworm from the image input to the image input unit 200 using the learning model 300, and the learning model 300 If an object is detected according to the object detection result, the detected object may be classified into at least one of normal stool, bloody stool, mucous stool, and hookworm using the learning model 300.

The notification providing unit 500 may output a preset notification to the user terminal 20 based on at least one of the detection result and classification result of the learning model 300 performed by the object detection and classification unit 400.

The method by which the notification providing unit 500 provides a preset notification to the user terminal 20 is to output at least one of the preset text and interface on the screen unit 22 of the user terminal 20 or to emit a preset notification sound. It may be at least one of those output through the sound output unit 23 of the user terminal 20, but is not limited thereto.

The notification providing unit 500 may be configured to output a notification using at least one of the screen unit 22 and the sound output unit 23 of the user terminal 20 in conjunction with software already installed on the user terminal 20. .

The notification provider 500 selects an object corresponding to at least one of normal stool, bloody stool, mucous stool, and hookworm input to the image input unit 200, according to the judgment result of the learning model 300 made in the object detection and classification unit 400. is detected, and if the detected object is classified as at least one of normal stool, bloody stool, mucous stool, and hookworm, a preset notification corresponding to the object classification result may be output to the user terminal 20.

The notification provider 500 provides normal stool, bloody stool, mucous stool, and hookworm, respectively, so that a preset notification corresponding to the classification result of the learning model 300 made in the object detection and classification unit 400 can be output to the user terminal 20. A notification corresponding to may be stored, and in this case, the notification may be at least one of a preset notification phrase, a preset notification sound, and a preset notification interface.

The preset notification stored in the notification providing unit 500 may be for informing of the status of the companion animal's stool.

For example, the notification provider 500 detects an object in an image input to the image input unit 200 according to the classification result of the learning model 300 made in the object detection and classification unit 400, and the detected object is In this case, a preset phrase to notify that hookworms are present in the pet's stool can be output on the user terminal 20.

The notification provider 500 detects a plurality of objects according to the detection result of the learning model 300 performed by the object detection and classification unit 400, and responds to the classification result when the plurality of detected objects are classified as different from each other. This may be such that a plurality of notifications are output to the user terminal 20.

For example, the notification provider 500 detects two objects according to the detection results of the learning model 300 performed by the object detection and classification unit 400, and the two detected objects are classified as bloody stool and mucous stool, respectively. In this case, a notification corresponding to bloody stool and a notification corresponding to mucous stool may be output together on the user terminal 20.

The notification provider 500 detects an object according to the detection result of the learning model 300 performed by the object detection and classification unit 400, and classifies the detected object as corresponding to at least two of bloody stool, mucous stool, and hookworm. If so, a plurality of notifications corresponding to the classification results may be output to the user terminal 20.

For example, the notification provider 500 determines that one object is detected according to the detection result of the learning model 300 performed by the object detection and classification unit 400, and that the one detected object corresponds to bloody stool and mucous stool. When classified, a notification corresponding to bloody stool and a notification corresponding to mucous stool may be output together on the user terminal 20.

The user terminal 20 through which the notification provider 500 outputs a notification may be a terminal that transmits an image to the image input unit 200.

The communication unit 600 may be connected to the terminal communication unit 24 of the user terminal 20 using an external communication network.

The communication unit 600 is connected to the terminal communication unit 24 of the user terminal 20 and can receive images input to the input unit 21 according to the user's input from the user terminal 20.

The notification providing unit 500 may transmit a signal to the user terminal 20 through the communication unit 70 so that a preset notification is output to the user terminal 20.

Meanwhile, in general, deep learning learning models for detecting objects in images require learning a large amount of images to detect objects, so the resolution of images included in the training data is often limited, for example. Images with relatively low resolution, such as 224x224 and 640x480, can be used as training data for deep learning models. In the case of hookworms, which can be present in the feces of companion animals, the size is relatively very small, so when learning a learning model to detect objects corresponding to hookworms using relatively low-resolution images such as 224x224 or 640x480, the number of pixels is limited. This may result in poor learning and detection performance.

When learning a learning model to detect an object corresponding to the above-described hookworm, in order to prevent problems with poor learning and detection performance due to the limited number of pixels, pet health status analysis using feces images according to an embodiment of the present invention The system may further include an image extraction unit 700 and an image preprocessing unit 800.

The database 100 is obtained by photographing the feces of a companion animal, and can pre-store learning images containing labeling data for a preset location area where an object corresponding to hookworm is located.

The database 100 is acquired by photographing the feces of a companion animal, and includes a preset coordinate area within which an area containing objects corresponding to hookworms is located, and labeling data including annotations corresponding to the coordinate area. A learning image containing may be pre-stored, and in this case, the annotation corresponding to the preset coordinate area in which the area containing the object corresponding to the hookworm is located may correspond to the hookworm.

The image extraction unit 700 extracts a patch image corresponding to a preset location area where a preset object corresponding to hookworm is located from the learning image stored in the database 100 based on the labeling data of the learning image previously stored in the database 100. can be extracted.

The image extraction unit 700 may extract a patch image from a learning image including labeling data including an annotation corresponding to a preset location area where a preset object is located and a hookworm among the learning images stored in the database 100. .

The image extraction unit 700 can extract a preset coordinate area of the learning image including a preset location area where a preset object is located and labeling data including annotations corresponding to hookworms as a patch image, and extract a patch image from a database (100 ), the image corresponding to the inside of the coordinate area included in the labeling data of the learning image stored in can be extracted as a patch image.

The image pre-processing unit 800 determines whether to normalize the size of the patch image based on at least one of the size and clarity of the patch image, and if it is determined that normalization of the size is necessary according to the determination result, it adjusts the size of the patch image. It may be normalized and the normalized patch image may be stored in the database 100.

The image pre-processing unit 800 includes the size of the patch image extracted from the image extraction unit 700, a size parameter calculated based on a preset mathematical equation, and the sharpness of the patch image extracted from the image extraction unit 700 and a preset Based on at least one of the clarity parameters calculated based on the equation, it is determined whether normalization of the size of the patch image extracted from the image extraction unit 700 is necessary, and the size of the predetermined patch image is determined according to the determination result. If it is determined that normalization is necessary, the size of a certain patch image may be normalized, and the normalized patch image may be stored in the database 100.

The image pre-processing unit 800 calculates a size parameter based on the size of a predetermined patch image and a preset mathematical equation, and if the calculated size parameter is greater than a preset threshold, it determines that normalization of the size of the predetermined patch image is necessary. , the size of the predetermined patch image may be normalized and the normalized predetermined patch image may be stored in the database 100.

The image pre-processing unit 800 determines that the size of the patch image extracted from the image extraction unit 700 and the size parameter calculated based on the preset mathematical equation are smaller than the preset threshold value, and the sharpness of the predetermined patch image and the preset mathematical equation If the sharpness parameter calculated based on is smaller than the preset threshold, it is determined that normalization of the size of the predetermined patch image is necessary, the size of the predetermined patch image is normalized, and the normalized predetermined patch image is stored in the database 100. It may be to save it.

The image pre-processing unit 800 may include an image classification unit 810 and a size normalization unit 820.

The image classification unit 810 includes the size of the patch image extracted from the image extraction unit 700, a size parameter calculated based on a preset mathematical equation, and the sharpness of the patch image extracted from the image extraction unit 700 and a preset Patch images that require normalization of size among patch images extracted from the image extraction unit 700 may be classified based on at least one of the sharpness parameters calculated based on a mathematical equation. When the image classification unit 810 calculates the size parameter, the size of the patch image based on it may be at least one of the horizontal length and vertical length of the patch image.

The image classification unit 810 calculates the size parameter based on the size of the patch image extracted from the image extraction unit 700 and the preset Equation 1, and selects a patch image with a size parameter greater than the preset first threshold. It may be classified as a patch image that requires normalization of size.

In Equation 1, P ₁ is a size parameter, and min(R _patch , C _patch ) may be the smaller of the horizontal and vertical lengths of the patch image extracted from the image extraction unit 700.

Meanwhile, in the size normalization unit 820, the size of the patch image is normalized to a preset size, and in Equation 1, min(R _resizer , C _resizer ) may be the smaller of the horizontal and vertical lengths of the normalized patch image. .

In other words, the image classification unit 810 may determine that size normalization is necessary for patch images that satisfy Equation 2 below among the patch images extracted from the image extraction unit 700 and classify them.

In Equation 2, min(R _patch , C _patch ) is the smaller of the horizontal and vertical lengths of the patch image extracted from the image extraction unit 700, and min(R _resizer , C _resizer ) is the smaller of the horizontal and vertical lengths of the patch image extracted from the image extraction unit 700. It is the smaller of the horizontal and vertical lengths of the normalized patch image, and Th ₁ is the preset first threshold.

The smaller of the horizontal and vertical lengths of the patch image that satisfies Equation 2 may be relatively smaller than the smaller of the horizontal and vertical lengths of the patch image normalized in the size normalization unit 820.

In other words, that the patch image satisfies Equation 2 means that the image extraction unit 700 extracts the patch image that satisfies Equation 2 using the learning image that is included in the labeling data of the learning image when training the learning model 300. This may mean that the size of the coordinate area is too small and the learning efficiency of the learning model 300 may be reduced.

In order to prevent the learning efficiency of the learning model 300 from decreasing, the image classification unit 810 determines that patch images that satisfy Equation 2 require size normalization and classifies them, and the size normalization unit 820 The size of the patch image classified by the image classification unit 810 can be enlarged to a preset size so that the learning model 300 can be efficiently learned.

The image classification unit 810 determines the sharpness parameter based on the average brightness value in the preset coordinate area of the patch image extracted from the image extraction unit 700 and a preset Laplacian operator for edge extraction. The patch image may be calculated and classified as a patch image that requires normalization for size, having a sharpness parameter smaller than the preset second threshold.

The image classification unit 810 can position the patch image extracted from the image extraction unit 700 in a preset XY coordinate system and calculate the sharpness parameter using Equation 3.

In Equation 3, SH may be a sharpness parameter, and LP(x,y) may be a preset Laplacian filter for edge extraction expressed in Equation 4 below, extracted from the image extraction unit 700. It may be a value calculated when a convolution is performed by applying it to (x,y), a preset coordinate position on the patch image, and u _xy is obtained from the pixel located at (x,y), a preset coordinate position on the patch image. It may be the average brightness value of pixels located within a preset distance.

The image classification unit 810 performs a convolution operation by applying the Laplacian filter expressed in Equation 4 to (x, y), which is a preset coordinate position on the patch image extracted from the image extraction unit 700. The sharpness parameter can be calculated based on the results of performing the convolution operation and Equation 3.

By applying the Laplacian filter expressed in Equation 4 to an image in which a preset object exists and performing a convolution operation, the edge of the object can be extracted. The resolution of the image is relatively high, so the clearer the image, the LP ( x,y) can be calculated as a high value, and LP(x,y) can be calculated as a low value as the resolution of the image is relatively low and the image is blurry.

The image classification unit 810 calculates the average brightness value of pixels located within a preset distance from the pixel located at the preset coordinate position (x, y) on the patch image extracted from the image extraction unit 700 through a preset algorithm. can do.

The image classification unit 810 may determine that size normalization is required for patch images that satisfy Equation 5 below among the patch images extracted from the image extraction unit 700 and classify them.

In Equation 5, SH is the sharpness parameter calculated by Equation 3, and Th2 is the preset second threshold.

A patch image that satisfies Equation 5 may have a relatively lower resolution than an image with a sharpness parameter corresponding to a preset second threshold.

In other words, that the patch image satisfies Equation 5 means that the image extraction unit 700 extracts the patch image that satisfies Equation 5 using the learning image that is included in the labeling data of the learning image when training the learning model 300. As the clarity of the coordinate area is low, it may mean that the resolution is low and the learning efficiency of the learning model 300 may be reduced.

In order to prevent the learning efficiency of the learning model 300 from decreasing, the image classification unit 810 determines that patch images that satisfy Equation 5 require size normalization and classifies them, and the size normalization unit 820 The size of the patch image classified in the image classification unit 810 can be adjusted so that the learning model 300 can be learned efficiently.

The size normalization unit 820 may normalize the size of the patch images classified by the image classification unit 810 to a preset size and store the normalized patch images in the database 100.

The size normalization unit 820 uses a preset image bilinear interpolation algorithm to enlarge the size of the patch image classified in the image classification unit 810 to a preset size, thereby classifying the patch image in the image classification unit 810. This may be normalizing the size of the patch image.

When the size normalization unit 820 normalizes the size of the patch image classified by the image classification unit 810 by enlarging it to a preset size, the patch image can have a size sufficient for efficient learning of the learning model 300. there is.

The learning model 300 may be pre-trained to detect preset objects corresponding to hookworms by applying training data including learning images and patch images stored in the database 100, and the object detection and classification unit 400 may detect an object corresponding to a hookworm in an image input to the image input unit 200 using the learning model 300.

The learning model 300 may be pre-trained to detect objects corresponding to hookworms in images using a pre-set deep learning model. At this time, the pre-set deep learning model can generally detect objects in images. There is no limitation as long as it exists, and for example, it can be any one of YOLO, v3, SSD, and Faster RCCN.

The notification provider 500 outputs a preset notification to the user terminal 20 when an object corresponding to hookworm is detected in the image input to the image input unit 200 according to the detection result of the object detection and classification unit 400. It may be.

If an object corresponding to hookworm is detected in the image input to the image input unit 200 according to the detection result of the object detection and classification unit 400, the notification provider 500 sends a dewormer message to the user terminal 20 in the feces of the companion animal. This may be outputting a preset notification to notify of its existence.

The notification providing unit 500 may store a preset notification for notifying the presence of hookworms in the pet's stool. In this case, the notification may be at least one of a notification phrase, a notification sound, and a notification interface.

The above-described embodiments are merely illustrative, and various modifications thereof are possible for those skilled in the art.

Therefore, the true technical protection scope of the present invention should include not only the above embodiments but also other variously modified embodiments according to the technical spirit of the invention described in the claims below.

10: Pet health status analysis system,
100: database,
200: Image input unit,
300: Learning model,
400: Object detection and classification unit,
500: Notification provision department,
600: Department of Communications;
700: Image extraction unit,
800: Image preprocessing unit,
810: image classification unit, 820: size normalization unit,
20: user terminal,
21: input unit, 22: screen unit, 23: sound output unit, 24: terminal communication unit.

Claims (4)

By communicating with a user terminal that includes an input unit that receives input values according to the user's input, a screen unit that can display preset text and interfaces, and a sound output unit that can output sound,
A database that stores learning images obtained by photographing the stool of a companion animal and including labeling data for the location area of a preset object corresponding to at least one of normal stool, bloody stool, mucous stool, and hookworm;
An image input unit that receives images obtained by photographing the feces of a companion animal from the user terminal;
Training data including learning images stored in the database are applied to detect objects corresponding to at least one of normal stool, bloody stool, mucous stool, and hookworm in the image. When an object is detected, the detected object is classified into normal stool, bloody stool, mucous stool, and hookworm. A learning model that is pre-trained so that it can be classified into at least one of the following;
an object detection and classification unit that detects an object from an image input to the image input unit using the learning model and, when an object is detected, classifies the object into at least one of normal stool, bloody stool, mucous stool, and hookworm using the learning model; and
A notification providing unit configured to output a preset notification to the user terminal based on at least one of the object detection result and the object classification result of the object detection and classification unit.
Pet health status analysis system using human feces images. According to clause 1,
From the learning images stored in the database based on the preset location area where the object corresponding to the hookworm is located among the learning images stored in the database, and the labeling data of the learning image including labeling data including annotations corresponding to the location area. An image extraction unit that extracts a patch image corresponding to the location area of a preset object corresponding to hookworm; and
It is determined whether normalization of the size of the patch image is necessary based on at least one of the patch image size and clarity, and if it is determined that normalization of the size is necessary according to the determination result, the size of the patch image is normalized and stored in the database. It further includes an image pre-processing unit for storing in,
The learning model is
Pre-trained to detect preset objects corresponding to hookworms by applying training data including patch images normalized in the image pre-processing unit and stored in the database.
Pet health status analysis system using human feces images. According to clause 2,
The image preprocessing unit
An image classification unit that determines and classifies the patch images extracted from the image extraction unit as requiring normalization of size based on at least one of the size and clarity of the patch images extracted from the image extraction unit; and
A size normalization unit that normalizes the patch images classified by the image classification unit to have a preset size and stores the normalized patch images in the database.
Pet health status analysis system using human feces images. According to clause 3,
The image classification unit
Based on the size of the patch image extracted from the image extraction unit and a preset mathematical equation
Calculate size parameters and classify patch images whose calculated size parameters are larger than a preset threshold as patch images that require size normalization.
Pet health status analysis system using human feces images.

Images