Method and device for diagnosing eye disease of animal by lightening artificial intelligence diagnosis model

Abstract

The present invention relates to a method and device for diagnosing an eye disease of an animal by lightening an artificial intelligence diagnosis model. The method for diagnosing an eye disease of an animal by lightening an artificial intelligence diagnosis model comprises the steps in which: a data processing unit receives a diagnosis model and a constraint including a region of interest set by a user according to the type of eye disease; the data processing unit acquires a plurality of lightened diagnosis models according to the constraint by applying, to the diagnosis model, lightening methods for lightening the diagnosis model on the basis of the constraint; the data processing unit acquires captured image data obtained by imaging the eyes of the animal; the data processing unit acquires, from the captured image data, pieces of analysis image data including a region of interest for each type of eye disease; the data processing unit trains, with the pieces of analysis image data, at least one among the plurality of lightened diagnosis models; and the data processing unit generates diagnosis data for the animal by using the at least one trained lightened diagnosis model among the plurality of lightened diagnosis models.

Description

Method and device for diagnosing animal eye diseases using lightweighting of artificial intelligence diagnostic models

[0001] The present invention relates to a method and apparatus for diagnosing eye diseases in animals using lightweighting of an artificial intelligence diagnostic model, and more specifically, to a method and apparatus for diagnosing eye diseases in animals using lightweighting of an artificial intelligence diagnostic model that allows a user to self-diagnose eye diseases in animals.

[0002] As the population ages and the number of single-person households increases, humans are becoming increasingly self-centered and emotionally desolate. Consequently, the number of people who view pets as family members or companions is increasing, and the pet market is also growing steadily.

[0003] Since the National Institute of Animal Science of the Rural Development Administration announced in 2008 that there are 2 million households raising pets in Korea, raising approximately 5 million animals, the number of households raising pets has been continuously increasing.

[0004] In cases where unusual symptoms not usually seen in pets occur, most people have taken their pets to an animal hospital for treatment or resolved the symptoms based on information obtained from people around them, the internet, or phone calls.

[0005] However, since information obtained from acquaintances or via the internet and phone often contains inaccuracies, it can be difficult to receive treatment. Even when visiting a hospital in person, there are frequently cases where proper service is not provided to customers due to long waiting times and a heavy workload.

[0006] Therefore, there is a need for individuals to conveniently diagnose their pets' diseases without the help of medical professionals and to establish a medical or treatment plan based on the diagnosis results.

[0007] The matters described as background technology are intended only to enhance understanding of the background of the present invention and should not be construed as an acknowledgment that they constitute prior art already known to those skilled in the art.

[0008] The present invention provides a method and apparatus for diagnosing animal eye diseases using lightweighting of an artificial intelligence diagnostic model capable of self-diagnosing animal eye diseases.

[0009] The problems that the present invention aims to solve are not limited to those mentioned above, and other unmentioned problems will be clearly understood by a person skilled in the art from the description below.

[0010] A method for diagnosing an eye disease in an animal using an artificial intelligence diagnostic model lightweighting according to an embodiment of the present invention is a method for diagnosing an eye disease in an animal performed by a data processing unit that is mounted on a user terminal and capable of diagnosing an eye disease from captured image data of an animal's eye, wherein the data processing unit receives a constraint including a diagnostic model and a region of interest according to a type of eye disease set by the user; the data processing unit applies a lightweighting technique to the diagnostic model to lightweight the diagnostic model based on the constraint to obtain a plurality of lightweight diagnostic models according to the constraint; the data processing unit obtains captured image data of an animal's eye; the data processing unit obtains analysis image data including a region of interest for each type of eye disease from the captured image data; the data processing unit trains at least one of the plurality of lightweight diagnostic models using the analysis image data; and the data processing unit generates diagnostic data of the animal using at least one of the trained plurality of lightweight diagnostic models.

[0011] The above constraints may further include at least one of the type of animal, the type of user terminal, accuracy, latency, and energy consumption.

[0012] The process of acquiring the plurality of lightweight diagnostic models may include: a process of determining combinations of lightweighting techniques to be applied to the diagnostic models according to the constraints; a process of generating a plurality of lightweight models by applying the combinations of lightweighting techniques to the diagnostic models; a process of acquiring performance for each of the plurality of lightweight models; and a process of determining the lightweight model with the best performance among the plurality of lightweight models according to the constraints as the lightweight diagnostic model.

[0013] The above lightweighting technique may include at least one of pruning, knowledge distillation, NAS (Neural Architecture Search), and quantization techniques.

[0014] The process of determining the combinations of the above-mentioned lightweighting techniques may include the process of determining the types and application order of the above-mentioned lightweighting techniques.

[0015] When the animal's eye is detected by the above-mentioned imaging unit, the control unit can recognize the animal's eye as an object and control the operation of the imaging unit to automatically photograph the animal's eye and acquire the captured image data.

[0016] An animal eye disease diagnosis device using an artificial intelligence diagnostic model lightweighting according to an embodiment of the present invention comprises: a shooting unit capable of capturing an animal's eye to acquire captured image data; and a data processing unit capable of acquiring a plurality of analysis image data including a region of interest according to the type of eye disease from the captured image data, and generating diagnosis result data from the plurality of analysis image data using a learned lightweighting diagnostic model. The data processing unit receives the diagnostic model and the region of interest as constraints, and can combine at least one lightweighting technique based on the constraints and apply it to the diagnostic model.

[0017] The above lightweighting technique includes at least one of pruning, knowledge distillation, NAS (Neural Architecture Search), and quantization techniques, and the data processing unit can generate the lightweighting diagnostic model by combining the type and application order of the lightweighting technique according to the region of interest.

[0018] The apparatus further includes a control unit capable of automatically controlling the operation of the above-mentioned shooting unit, and the control unit can control the shooting unit to automatically photograph the animal's eye when the animal's eye is recognized by the shooting unit.

[0019] Other specific details of the present invention are included in the detailed description and drawings.

[0020] According to the present invention, an image of an animal's eye, such as a pet, can be acquired using a user terminal, and an eye disease can be self-diagnosed from the acquired image. That is, an eye disease of an animal can be diagnosed using a user terminal equipped with a lightweight artificial intelligence model capable of self-diagnosing the animal's eye disease. Therefore, the user can quickly and easily diagnose or predict an eye disease of an animal, such as a pet, without having to visit an animal hospital in person with the animal.

[0021] Furthermore, since it can overcome communication and cost issues associated with conventional server-based pet diagnostic services, it offers the advantage of reducing communication and diagnostic costs. Additionally, by utilizing a lightweight artificial intelligence model, the inference time for diagnosis can be shortened, providing the advantage of rapidly diagnosing eye diseases in pets.

[0022] The effects of the present invention are not limited to those mentioned above, and other unmentioned effects will be clearly understood by a person skilled in the art from the description below.

[0023] FIG. 1 is a schematic diagram showing an animal eye disease diagnosis system using lightweighting of an artificial intelligence diagnostic model according to an embodiment of the present invention.

[0024] FIG. 2 is a block diagram of an animal eye disease diagnosis device using lightweighting of an artificial intelligence diagnostic model according to an embodiment of the present invention.

[0025] FIG. 3 is a flowchart showing a method for diagnosing eye diseases in animals using lightweighting of an artificial intelligence diagnostic model according to an embodiment of the present invention.

[0026] FIG. 4 is a diagram illustrating the process of extracting analysis image data in a method for diagnosing eye diseases in animals using lightweighting of an artificial intelligence diagnostic model according to an embodiment of the present invention.

[0027] FIGS. 5 and 6 are drawings for explaining the process of diagnosing an eye disease in an animal using a lightweight artificial intelligence diagnostic model according to an embodiment of the present invention.

[0028] The advantages and features of the present invention and the methods for achieving them will become clear by referring to the embodiments described in detail below together with the appendix. However, the present invention is not limited to the embodiments disclosed below but may be implemented in various different forms. These embodiments are provided merely to ensure that the disclosure of the present invention is complete and to fully inform those skilled in the art of the scope of the present invention, and the present invention is defined only by the scope of the claims.

[0029] The terms used in this specification are for describing embodiments and are not intended to limit the invention. In this specification, the singular form includes the plural form unless specifically stated otherwise in the text. The terms "comprises" and / or "comprising" used in this specification do not exclude the presence or addition of one or more other components in addition to the components mentioned. Throughout the specification, the same reference numerals refer to the same components, and "and / or" includes each of the mentioned components and all combinations of one or more. Although terms such as "first," "second," etc., are used to describe various components, these components are not limited by these terms. These terms are used merely to distinguish one component from another. Therefore, the first component mentioned below may be the second component within the technical scope of the invention.

[0030] Unless otherwise defined, all terms used herein (including technical and scientific terms) may be used in a meaning commonly understood by those skilled in the art to which the present invention pertains. Additionally, terms defined in commonly used dictionaries are not to be interpreted ideally or excessively unless explicitly and specifically defined otherwise.

[0031] First, in this embodiment, the method and device for diagnosing eye diseases in animals using lightweight artificial intelligence diagnostic models are described as diagnosing the type of disease or illness that the animal has by photographing the eyes of a pet, such as a dog or a cat, using a user terminal owned by the user; however, the animal is not limited to pets and may include various animals such as vertebrates like mammals, birds, reptiles, amphibians, and fish, and invertebrates like arthropods and mollusks. Additionally, the user may be a guardian raising a pet, but may also include an operator of a facility handling animals, such as a pet shop, zoo, or animal shelter, or an employee of such a facility.

[0032] Hereinafter, embodiments of the present invention will be described in detail with reference to the attached drawings.

[0033] FIG. 1 is a schematic diagram showing an animal eye disease diagnosis system using artificial intelligence diagnostic model lightweighting according to an embodiment of the present invention, and FIG. 2 is a block diagram of an animal eye disease diagnosis device using artificial intelligence diagnostic model lightweighting according to an embodiment of the present invention.

[0034] Referring to FIG. 1, an animal eye disease diagnosis system using an artificial intelligence diagnostic model lightweighting according to an embodiment of the present invention may include a user terminal (100) and a management server (200). Additionally, an animal eye disease diagnosis system according to an embodiment of the present invention may further include an animal hospital server (300) capable of transmitting and receiving information with the user terminal (100) and the management server (200).

[0035] Here, the user terminal (10), the management server (20), and the animal hospital server (300) can synchronize and transmit and receive data in real time using the communication network (400). The communication network (400) may support various long-distance communication methods, for example, Wireless LAN (WLAN), DLNA (Digital Living Network Alliance), Wibro (Wireless Broadband), Wimax (World Interoperability for Microwave Access), GSM (Global System for Mobile communication), CDMA (Code Division Multi Access), CDMA2000 (Code Division Multi Access 2000), EV-DO (Enhanced Voice-Data Optimized or Enhanced Voice-Data Only), WCDMA (Wideband CDMA), HSDPA (High Speed ​​Downlink Packet Access), HSUPA (High Speed ​​Uplink Packet Access), IEEE 802.16, Long Term Evolution (LTE), LTEA (Long Term Evolution-Advanced), Wireless Mobile Broadband Service (WMBS), BLE (Bluetooth Low Energy), Zigbee, RF (Radio Various communication methods such as Frequency and LoRa (Long Range) may be applied, but are not limited to these; various widely known wireless or mobile communication methods may also be applied.

[0036] First, in this embodiment, the user terminal (100) is used to photograph the eyes of a pet (A) to obtain an image, and the presence or absence of an eye disease in the pet and the type of eye disease are diagnosed from the obtained image, but this is not limited thereto. Such a user terminal (100) may be an eye disease diagnostic device capable of diagnosing an eye disease in an animal.

[0037] A user terminal (100) is a portable terminal possessed by the guardian of a pet (A) and can operate using an application program (or application) in the present disclosure, and such an application program can be downloaded from an external server or a management server (200) via wireless communication. For example, the user terminal (100) may include various terminals such as a smartphone, PDA (Personal Digital Assistant), tablet, wearable device (e.g., watch-type terminal (Smartwatch), glass-type terminal (Smart Glass), HMD (Head Mounted Display), etc.) and various IoT (Internet of Things) terminals, but is not limited thereto.

[0038] A user terminal (100) can recognize the eyes on the face of a pet (A) and acquire captured image data centered on the eyes, such as the eyeball. Additionally, the user terminal (100) can determine whether the pet has an eye disease from the captured image data using an artificial intelligence model installed therein, such as a lightweight artificial intelligence model, and if an eye disease is present, predict or diagnose the type of eye disease.

[0039] Referring to FIG. 2, the eye disease diagnostic device, i.e., the user terminal (100), may include a shooting unit (110), a communication unit (120), a storage unit (130), a data processing unit (140), and a control unit (170). Additionally, the user terminal (100) may further include an input unit (150) and an output unit (160).

[0040] The shooting unit (110) can use a camera (not shown) installed on the user terminal (100) to recognize the eyes or eyeballs on the face of the pet (A) and obtain captured image data centered on the eyeballs. The captured image data may include information about photos or videos of the pet's condition requiring care. For example, the captured image data may include photos or videos of the pet's eye condition, but is not limited thereto.

[0041] The shooting unit (110) can be operated automatically through the control of the control unit (170) described later, and when an eye or eyeball is recognized on the face of the pet (A), the eye of the pet (A) can be automatically photographed to obtain the captured image data.

[0042] The communication unit (120) can transmit and receive various data with the management server (200) and the animal hospital server (300). According to an embodiment, the communication unit (120) can receive an application program or application from the management server (200). Additionally, the communication unit (120) can transmit result data and diagnosis data regarding the diagnosis of eye diseases of pets to the animal hospital server (300). Furthermore, the communication unit (120) can transmit a reservation request signal to the animal hospital server (300) and receive a reservation completion signal from the animal hospital server (300).

[0043] The storage unit (130) can store a number of application programs (or applications) running on the user terminal (100), data for the operation of the user terminal (100), and commands. At least some of these application programs can be downloaded from the management server (200) via wireless communication.

[0044] Additionally, the storage unit (130) can store a diagnostic model, various training data for training the diagnostic model, and diagnostic result data generated as a result of training. Here, the training data may include various eye disease data and may be stored as image data in the storage unit (130). According to an embodiment, the diagnostic model may include a DCIN algorithm developed based on a CNN algorithm.

[0045] Additionally, the storage unit (130) can store a diagnostic model and a lightweighting technique (lightweighting algorithm) for lightweighting the diagnostic model. Here, well-known lightweighting techniques such as pruning, knowledge distillation, NAS (Neural Architecture Search), and quantization techniques may be utilized.

[0046] The data processing unit (140) can train a diagnostic model using the training data stored in the storage unit (130), and can diagnose an eye disease in a pet from captured image data, such as a photo or video of the pet's eyes, using the trained diagnostic model.

[0047] The data processing unit (140) can automatically adjust brightness, clarity, etc. based on the image data obtained through the shooting unit (110), taking into account the surrounding environment, shaking, blinking, and whether the pupil is captured.

[0048] The data processing unit (140) can extract an analysisable image from the captured image data by considering the surrounding environment, shaking, blinking, whether the pupil is captured, etc., and can automatically correct the brightness, sharpness, etc. of the extracted analysis image. In addition, the data processing unit (140) can extract multiple analysis image data by region of interest from the corrected analysis image as shown in FIG. 5. For example, the multiple analysis image data may each be image data for a specific area for analyzing multiple diseases, such as the pupil, the white of the eye, and the eyelid.

[0049] The data processing unit (140) can generate standard data by training a diagnostic model using the training data stored in the storage unit (130). For example, the training data may include basic ophthalmic condition data and basic ophthalmic result data, and the data processing unit (140) can generate standard data by matching the basic ophthalmic condition data and the basic ophthalmic result data. Here, the basic ophthalmic condition data may include basic data including the type, age, gender, and neutering status of the pet, and image data of the pet's eyes. At this time, the image data may include partial images extracted by region of interest according to the type of eye disease. For example, the image data may be images generated by extracting the eyeball, the white of the eye, and the eyelid area. This is to diagnose by identifying the symptoms that occur in each area, as the types and symptoms of eye diseases that occur in each region of interest differ.

[0050] Basic ophthalmic result data may include basic diagnostic data and basic judgment data. Basic diagnostic data is data used to diagnose the presence or absence of ophthalmic diseases using basic ophthalmic imaging information, and basic judgment data may include data used to determine the disease-specific progression stages of ophthalmic diseases based on the basic diagnostic data. Standard data may be data generated by matching the basic ophthalmic status data and basic ophthalmic result data of a companion animal.

[0051] The data processing unit (140) may generate standard data using machine learning techniques such as Random Forest, Support Vector Machine, or deep learning. Specifically, the data processing unit (140) may generate standard data by verifying suitability through iterative learning of basic ophthalmic condition information and basic ophthalmic result data based on a Convolutional Neural Network (CNN) algorithm. At this time, the process of verifying the standard data may cross-verify suitability by veterinarians and researchers of the commissioned research period, for example, at least three specialists, but is not limited thereto.

[0052] Meanwhile, the data processing unit (140) can generate a lightweight diagnostic model by applying at least one lightweighting technique to the diagnostic model when training the diagnostic model, and can generate standard data using the generated lightweight diagnostic model. The lightweighting technique may include at least one of the NAS technique (Neural Architecture Search), pruning technique, knowledge distillation technique, and quantization technique.

[0053] Neural Architecture Search (NAS) is a technique that can create lightweight deep learning models by searching for the optimal model structure, thereby saving time and resources.

[0054] Pruning is a technique that reduces the size of a deep learning model and lowers computational costs by removing layers with low learning contribution. Pruning utilizes the fact that a significant number of parameters (weights) in a deep learning model do not contribute much to actual inference, thereby removing unnecessary connections to achieve lightweighting without degrading model performance.

[0055] Knowledge Distillation is a technique for transferring knowledge from a complex large model (teacher model) to a simple small model (student model). By improving efficiency based on the learning results of the teacher model, the student model can simultaneously achieve performance and lightweight design.

[0056] Quantization is a technique that reduces computation and memory usage by representing weights used in a model with fewer bits, such as 8 bits or 4 bits, instead of high precision of 32 bits or more, thereby consuming much less hardware resources and making diagnostic models lighter.

[0057] According to an embodiment, the data processing unit (140) can generate a lightweight diagnostic model by combining lightweighting techniques according to the region of interest. Then, training data can be trained using the lightweight diagnostic model, and the performance of the lightweight diagnostic model can be evaluated. Then, training data for each region of interest can be trained using the lightweight diagnostic model with the best performance for each region of interest, and eye diseases of animals can be diagnosed from analysis images for each region of interest using this.

[0058] The reason for generating lightweight diagnostic models for each region of interest by combining lightweighting techniques according to the region of interest in this manner is as follows.

[0059] Types of eye diseases may include dry eye, ulcerative keratitis, non-ulcerative keratitis, conjunctivitis, glaucoma, uveitis, epiphora, etc., and the areas where the disease occurs differ depending on the type of eye disease. In addition, since the symptoms occurring in each area differ, the analysis image data for each area of ​​interest may have different characteristics. For example, analysis image data extracted with the pupil area as the area of ​​interest may have wounds or foreign objects in the pupil, and analysis image data extracted with the sclera area as the area of ​​interest may have blood vessels in the sclera area. Also, analysis image data extracted with the eyelid area as the area of ​​interest may have eyelashes or protrusions. As such, since the analysis image data for each area of ​​interest has different characteristics, the data processing unit (140) can generate result data including the presence or absence of disease and the type of disease that occurred by using different lightweight diagnostic models based on the analysis image data for each area of ​​interest.

[0060] In addition, the data processing unit (140) can generate diagnostic data including management information for managing pets with diseases, animal hospital linkage information, etc., based on the result data.

[0061] The data processing unit (140) can transmit and receive reservation management information between the user terminal (100) and the animal hospital server (300). For example, the data processing unit (140) can generate a reservation request signal for visiting the animal hospital and transmit it to the animal hospital server (300) via the communication unit (120), and can receive a reservation completion signal or a reservation acceptance signal for the reservation request signal from the animal hospital server (300) via the communication unit (120). In addition, the data processing unit (140) can transmit result data and diagnosis data to the animal hospital server (300) for the pet's hospital visit.

[0062] The input unit (150) may be provided to input information about the animal, such as a pet, for the diagnosis of the animal. The input unit (150) may be provided as a keypad, a touchscreen, etc., but is not limited thereto and may be various.

[0063] The output unit (160) can visually and audibly display the current operating status of the user terminal (100). Additionally, the output unit (160) may include a display capable of displaying symbols, characters, numbers, etc. on a screen according to the operating status, a lamp capable of displaying a color change or blinking, or a speaker capable of displaying audio.

[0064] The output unit (160) can output a guideline for photographing the pet's eyes so that the user can easily photograph the pet's eyes. For example, the output unit (160) can output a guideline in the form of multiple concentric circles centered on the pupil.

[0065] The control unit (170) can control the overall operation of the shooting unit (110), storage unit (130), data processing unit (140), input unit (150), and output unit (160). In particular, the control unit (170) can control the operation of the shooting unit (110) so that when the pet's eyes are recognized in the image obtained from the shooting unit (110), the pet's eyes are automatically photographed. Additionally, the control unit (170) can control the output unit (160) to output guidelines so that the user can accurately photograph the pet's eyes.

[0066] A user terminal (100) can generate result data and diagnostic data including the presence or absence of eye disease and the type of eye disease by comparing and analyzing the extracted analysis image data to analyze multiple diseases based on verified standard data obtained by repeatedly learning learning data obtained from multiple pets and analyzing the extracted analysis image data based on verified standard data and the shooting unit (110) of the user terminal (100). Accordingly, problems such as unnecessary hospital visits and neglect that may occur when judging the eye condition of a pet solely by visual images can be resolved. In addition, the user terminal (100) can diagnose eye diseases of pets while minimizing loss of accuracy by using a diagnostic model that applies a lightweighting technique.

[0067] Such a user terminal (100) can be implemented by hardware circuits (e.g., CMOS-based logic circuits), firmware, software, or a combination thereof. For example, it can be implemented by utilizing transistors, logic gates, and electronic circuits in the form of various electrical structures.

[0068]

[0069] Hereinafter, a method for diagnosing an animal eye disease using an animal eye disease diagnosis device utilizing lightweight artificial intelligence diagnostic models according to an embodiment of the present invention will be described.

[0070] FIG. 3 is a flowchart showing a method for diagnosing an eye disease in an animal using lightweighting of an artificial intelligence diagnostic model according to an embodiment of the present invention, FIG. 4 is a diagram explaining the process of extracting analysis image data in a method for diagnosing an eye disease in an animal using lightweighting of an artificial intelligence diagnostic model according to an embodiment of the present invention, and FIG. 5 and FIG. 6 are diagrams explaining the process of diagnosing an eye disease in an animal using lightweighting of an artificial intelligence diagnostic model according to an embodiment of the present invention.

[0071] Referring to FIG. 3, a method for diagnosing eye diseases in animals using artificial intelligence diagnostic model lightweighting according to an embodiment of the present invention is a self-diagnosis method for animals performed by a data processing unit (140) that is mounted on a user terminal (100) and capable of diagnosing eye diseases from captured image data of an animal's eye, wherein the data processing unit (140) receives a constraint including a diagnostic model and a region of interest according to the type of eye disease set by the user, a process (S110) in which the data processing unit (140) applies a plurality of lightweighting techniques to the diagnostic model to lightweight the diagnostic model based on the constraint to obtain a plurality of lightweight diagnostic models according to the constraint, a process (S120) in which the data processing unit (140) obtains captured image data of an animal's eye, a process (S130) in which the data processing unit (140) obtains analysis image data including a region of interest for each type of eye disease from the captured image data, and the data processing unit (140) uses the analysis image data to the plurality of lightweight diagnostic models The process may include a process of training at least one of the data processing unit (140) and a process of generating diagnostic data of the animal using at least one of the trained lightweight diagnostic models (S140, S150).

[0072] Here, animals may include pets, and pets can vary, such as dogs and cats.

[0073] First, a lightweight diagnostic model can be provided on a user terminal (100) to self-diagnose eye diseases of pets.

[0074] To this end, the user terminal (100) may receive and store a necessary application or application from a management server (200) that provides a service to self-diagnose a pet through a communication unit (120). At this time, the application or application may include a diagnostic model, a lightweighting technique (algorithm), and training data for training the diagnostic model.

[0075] A data processing unit (140) installed in a user terminal (100) can generate standard data by training a diagnostic model using training data. For example, the training data may include basic ophthalmic condition data and basic ophthalmic result data, and the data processing unit (140) can generate standard data by matching the basic ophthalmic condition data and the basic ophthalmic result data with each other. In this case, the training data may include partial images extracted by region of interest according to the type of eye disease from image data of a pet's eye.

[0076] The data processing unit (140) can combine lightweighting techniques for each region of interest and apply them to a diagnostic model to generate multiple lightweight diagnostic models. At this time, the data processing unit (140) receives the region of interest according to the type of eye disease set by the user as a constraint and can combine lightweighting techniques based on this. For example, if the region of interest is the pupil area, the data processing unit (140) can combine pruning techniques and quantization techniques and apply them to the diagnostic model. Alternatively, if the region of interest is the eyelid area, it can apply NAS techniques and knowledge distillation techniques to the diagnostic model. At this time, the data processing unit (140) can apply the types and order of lightweighting techniques to the diagnostic model by varying them for each region of interest.

[0077] Then, the generated lightweight diagnostic model can be used to train the learning data and evaluate the performance of the model. Subsequently, the lightweight diagnostic model exhibiting the best performance for each region of interest can be selected as the model for diagnosis in that region.

[0078] When the eyes of a pet are captured through the shooting unit (110) and the captured image data is obtained, the data processing unit (140) can extract an analysis image that can be analyzed based on the captured image data, taking into account the surrounding environment, shaking, blinking, and whether the pupil is captured, and can perform preprocessing to automatically adjust brightness, sharpness, etc. for the extracted analysis image.

[0079] Subsequently, as illustrated in FIG. 4, multiple analysis image data can be extracted for each region of interest from the analysis image. For example, the multiple analysis image data can each include a first analysis image data including the pupil area, a second analysis image data including the sclera area, and a third analysis image data including the eyelid area.

[0080] And, as illustrated in FIGS. 5 and 6, the data processing unit (140) can generate a first lightweight diagnostic model by combining a lightweighting technique with the first analysis image data and generate a first result data from the first analysis image data using the first lightweight diagnostic model. Additionally, it can generate a second lightweight diagnostic model by combining a lightweighting technique with the second analysis image data and generate a second result data from the second analysis image data using the second lightweight diagnostic model. Furthermore, it can generate a third lightweight diagnostic model by combining a lightweighting technique with the third analysis image data and generate a first result data from the third analysis image data using the third lightweight diagnostic model.

[0081] Here, it is explained that the first to third analysis image data are analyzed through different lightweighting diagnostic models, but at least two analysis image data may be analyzed through the same lightweighting diagnostic model.

[0082] When result data for analysis image data for each area of ​​interest is generated, the data processing unit (140) can generate diagnostic data for managing diseases of pets based on this.

[0083] The data processing unit (140) can generate a reservation request signal for visiting an animal hospital based on diagnostic data and transmit it to the animal hospital server (300) through the communication unit (120). Then, it can receive a reservation acceptance signal or a reservation completion signal from the animal hospital server (300) and output it to the output unit (160).

[0084] The steps of the method or algorithm described in connection with embodiments of the present invention may be implemented directly in hardware, implemented as a software module executed by hardware, or implemented by a combination thereof. The software module may reside in RAM (Random Access Memory), ROM (Read Only Memory), EPROM (Erasable Programmable ROM), EEPROM (Electrically Erasable Programmable ROM), Flash Memory, a hard disk, a removable disk, a CD-ROM, or any form of computer-readable recording medium well known in the art to which the present invention belongs.

[0085] Although embodiments of the present invention have been described above with reference to the attached drawings, those skilled in the art will understand that the present invention may be implemented in other specific forms without altering its technical concept or essential features. Therefore, the embodiments described above should be understood as illustrative in all respects and not restrictive.

[0086] 100: User terminal

[0087] 200: Management Server

[0088] 300: Animal Hospital Server

Claims

1. A method for diagnosing eye diseases in animals using lightweighting of an artificial intelligence diagnostic model, performed by a data processing unit that is installed on a user terminal and capable of diagnosing eye diseases from captured image data of an animal's eye, A process in which the above data processing unit receives input of constraints including a diagnostic model and a region of interest according to the type of eye disease set by the user; A process in which the data processing unit applies a lightweighting technique to the diagnostic model to lightweight the diagnostic model based on the constraints, thereby obtaining a plurality of lightweight diagnostic models according to the constraints; The process of the above data processing unit acquiring captured image data of an animal's eye; A process in which the above data processing unit obtains analysis image data including regions of interest for each type of eye disease from the above captured image data; The process of the data processing unit training at least one of the plurality of lightweight diagnostic models with the analysis image data; and A method for diagnosing eye diseases in animals using artificial intelligence diagnostic model lightweighting, comprising the process of generating diagnostic data of the animal using at least one of a plurality of trained lightweight diagnostic models by the data processing unit.

2. In Claim 1, The above constraint is a method for diagnosing eye diseases in animals using an artificial intelligence diagnostic model lightweighting, further comprising at least one of the type of animal, type of user terminal, accuracy, latency, and energy consumption.

3. In Claim 1, The process of acquiring the above-mentioned multiple lightweight diagnostic models is, A process of determining combinations of lightweighting techniques to be applied to the diagnostic model according to the above constraints; A process of generating multiple lightweight models by applying combinations of the above lightweighting techniques to the above diagnostic model; A process of acquiring performance for each of the above plurality of lightweight models; and A method for diagnosing eye diseases in animals using artificial intelligence diagnostic model lightweighting, comprising: a process of selecting the lightweight model with the best performance among the plurality of lightweight models according to the constraints as the lightweight diagnostic model.

4. In claim 1 or 3, The above lightweighting technique is a method for diagnosing eye diseases in animals using an artificial intelligence diagnostic model lightweighting technique that includes at least one of pruning, knowledge distillation, NAS (Neural Architecture Search), and quantization techniques.

5. In Claim 4, A method for diagnosing eye diseases in animals using artificial intelligence diagnostic model lightweighting, wherein the process of determining the combinations of the above lightweighting techniques includes the process of determining the types and application order of the above lightweighting techniques.

6. In Claim 4, The process of acquiring the above-mentioned captured image data includes a process in which a control unit mounted on the user terminal controls the operation of the capturing unit of the user terminal. A method for diagnosing eye diseases in animals using a lightweight artificial intelligence diagnostic model, wherein when the eye of the animal is detected by the above-mentioned imaging unit, the control unit recognizes the animal's eye as an object and controls the operation of the above-mentioned imaging unit to automatically photograph the animal's eye and obtain the above-mentioned image data.

7. A camera unit capable of capturing an animal's eye to acquire captured image data; and A data processing unit capable of acquiring a plurality of analysis image data including a region of interest according to the type of eye disease from the above-mentioned captured image data, and generating diagnosis result data from the plurality of analysis image data using a learned lightweight diagnostic model; The above data processing unit receives a diagnostic model and a region of interest as constraints, and combines at least one lightweighting technique based on the constraints and applies it to the diagnostic model, thereby diagnosing an eye disease in an animal using artificial intelligence diagnostic model lightweighting.

8. In Claim 7, The above lightweighting technique includes at least one of pruning, knowledge distillation, NAS (Neural Architecture Search), and quantization techniques, and The above data processing unit is an animal eye disease diagnosis device using artificial intelligence diagnostic model lightweighting, capable of generating the lightweight diagnostic model by combining the type and application order of the lightweighting technique according to the region of interest.

9. In Claim 7, It further includes a control unit capable of automatically controlling the operation of the above-mentioned shooting unit, and The above control unit is an animal eye disease diagnosis device using lightweight artificial intelligence diagnostic model, which can control the imaging unit to automatically photograph the animal's eye when the animal's eye is recognized by the imaging unit.

Images