KR102203320B1 - Learning dataset providing system based on artificial intelligence - Google Patents

Abstract

The present invention relates to an artificial intelligence-based learning data set providing system, the present invention is a user terminal; An administrator terminal for managing the user of the user terminal and the authority of the user; And a learning data providing server that is connected to each other through a wired/wireless communication network with the user terminal and the manager terminal, and collects image data and designates an object-specific area included in the image data to generate learning image data that separates each object. It characterized in that it includes;

Description

Artificial intelligence-based learning data set provision system {LEARNING DATASET PROVIDING SYSTEM BASED ON ARTIFICIAL INTELLIGENCE}

An embodiment of the present invention relates to an artificial intelligence-based learning data set providing system.

Machine learning is a field of artificial intelligence, which is the process of finding an appropriate answer to a new input based on the trained knowledge by training pre-prepared learning data on a computer.

In addition, object detection distinguishes the location and type of a specific object on an image, and in order to use machine learning for object detection, labeled learning data given both a question and a correct answer is required.

That is, in the case of object detection based on machine learning, the important thing along with feature extraction and learning algorithm is the collection of labeled learning data, and the more labeled learning data is provided, the more effectively learning can proceed.

However, it is difficult to generate a vast amount of labeled training data from thousands to tens of thousands.

Korean Registered Patent Publication No. 10-1879735 (July 12, 2018)

An object of the present invention is to provide an artificial intelligence-based learning data set providing system capable of generating artificial intelligence learning data by designating an object-specific area included in image data.

In addition, an object of the present invention is to provide an artificial intelligence-based learning data set providing system capable of allowing the artificial intelligence to generate new data by providing and learning learning data matching a specific search condition for an object. To do.

An artificial intelligence-based learning data set providing system according to the present invention for achieving the above object includes a user terminal; An administrator terminal for managing the user of the user terminal and the authority of the user; And a learning data providing server that is connected to each other through a wired/wireless communication network with the user terminal and the manager terminal, and collects image data and designates an object-specific area included in the image data to generate learning image data that separates each object. Including;, wherein the user terminal, a worker terminal for accessing the learning data providing server to designate an area for each object included in the image data; An inspector terminal accessing the learning data providing server to inspect an object-specific area of the image data designated by the worker terminal; And a searcher terminal connected to the learning data providing server to search for and receive learning image data to be provided to artificial intelligence through a search tool, wherein the inspector terminal comprises, for each object of the image data designated by the worker terminal A first inspector terminal that inspects the area and rejects or approves the operator terminal; And a second inspector terminal that rejects or approves the first inspector terminal by inspecting the object-specific area of the image data approved by the first inspector terminal, wherein the manager terminal includes the image data and the corresponding image data It is characterized by more managing the items of.

In addition, the learning data providing server includes a member registration unit for registering member information of a worker, an inspector, and a searcher, respectively, to authenticate a member; After collecting the image data, a learning data generation unit that designates an area for each object included in the corresponding image data to generate training image data in which each object is classified as image data to be learned by the artificial intelligence; A data management unit that manages the image data, items of image data, a user of a user terminal, and a user's authority, respectively; And a learning data detection unit that provides a search tool for the learning image data to provide learning image data to be learned by the artificial intelligence according to the search result.

In addition, the member registration unit, a worker member management unit for registering the member information of the worker to authenticate the member; An inspector member management unit for registering member information of the inspector to authenticate a member; And a searcher member management unit for registering member information of the searcher to authenticate a member, wherein the worker member management unit includes: a worker member registration module for registering member information of the worker; A worker member authentication module for authenticating member information of the worker; And a worker access permission module allowing access to the learning data providing server to a member of the worker who has been authenticated as a member, wherein the examiner member management unit includes: an examiner member registration module for registering member information of the examiner; An inspector member authentication module for authenticating the inspector's member information; And an inspector access permission module for allowing access to the learning data providing server to a member-authenticated inspector among the inspectors, wherein the searcher member management unit comprises: a searcher member registration module for registering member information of the searcher; A searcher member authentication module for authenticating member information of the searcher; And a searcher access permission module for allowing access to the learning data providing server to a searcher who has been authenticated as a member among the searchers.

In addition, the learning data generation unit may include an image data acquisition unit that collects image data and then classifies and acquires only image data that satisfies a specific object condition; An object area designation unit that designates an area for each object included in the image data acquired by the image data acquisition unit; An object area inspection unit that inspects an object-specific area of the image data designated by the object area designation unit; And a learning image data generation unit that generates corresponding image data as training image data when the object area inspection unit approves the area for each object.

In addition, the image data acquisition unit may include an image data collection module for collecting specific image data using a specific keyword on a website; An image data classification module for classifying image data by filtering only image data satisfying a specific object condition among image data collected by the image data collection module; An image data storage module temporarily storing the image data classified by the image data classification module; And a metadata generation module that generates metadata of image data classified by the image data classification module.

In addition, the object area designation unit displays an image data list to designate an object area among the image data on the operator terminal, and when the operator terminal selects specific image data from the image data list, an object-specific area of the image data You can provide a tool to specify.

In addition, the object area inspection unit may include: a first inspector inspection module for inspecting an object-specific area of the image data designated by the operator terminal; And a second inspector inspection module for inspecting an object-specific area of the image data approved by the first inspector terminal.

Further, the first inspector inspection module displays a list of workers on the first inspector terminal, and when the first inspector terminal selects a specific worker from the worker list, displays a list of work images of the corresponding worker, and 1 When the inspector terminal selects a specific work image from the work image list, a tool for checking an object-specific area of the work image may be provided.

In addition, the second inspector inspection module displays a list of workers on the second inspector terminal, and when the second inspector terminal selects a specific worker from the worker list, the inspection is completed by the first inspector terminal of the worker. When a work image list is displayed and the second inspector terminal selects a specific work image from the work image list, a tool for checking an object-specific area of the work image may be provided.

In addition, when the second inspector terminal checks an object-specific area of a specific work image of a specific worker in the worker list and finally approves the area for each object, the learning image data generation unit may You can create training image data that is a file.

In addition, the data management unit registers a folder in which the image data is to be finally stored, uploads and stores image data in the folder, and assigns image data to a worker terminal to designate an object-specific area of the image data. Management; An item management unit for managing an image type, which is an item of the image data, an object number and a meta number corresponding to the image type, respectively; A user management unit that registers and deletes a corresponding user based on user information of a user accessing the learning data providing server, or modifies and manages user information of a previously registered user; And an authority management unit that registers the user's authority to classify the user into an operator, an inspector, and a searcher, and then grants matching authority to each user.

In addition, the image data management unit may include a data upload module for registering the folder and uploading image data to the folder; And a data allocation module for allocating specific image data uploaded to the folder to a specific worker terminal.

In addition, the item management unit may include a type management module for registering and modifying an image type of the image data; An object management module for registering and modifying the number of objects of the image type; And a meta management module for registering and modifying the meta number of the image type.

In addition, the authority management unit, the authority registration module for registering the user's authority and classifying the user into an operator, an inspector, and a searcher; And an authorization module for granting respective rights to the operator, the inspector, and the searcher.

In addition, the learning data detection unit provides a search tool capable of searching for learning image data based on an image type of the learning image data, an area for each object, and metadata, and the metadata includes the expression, gender, and age of the object. It may include.

As described above, according to the artificial intelligence-based learning data set providing system according to the present invention, there is an effect of generating artificial intelligence learning data by designating a region for each object included in the image data.

In addition, the present invention has the effect of allowing the artificial intelligence to generate new data by providing and learning learning data matching a specific search condition for an object to the artificial intelligence.

1 is a block diagram of a system for providing learning data sets based on artificial intelligence according to an embodiment of the present invention.
2 is a block diagram of a learning data providing server according to the present invention.
3 is a configuration diagram of a member registration unit according to the present invention.
4 is a block diagram of a learning data generator according to the present invention.
5 is a configuration diagram of an image data acquisition unit according to the present invention.
6 and 7 are detailed implementation diagrams of an object area designating unit according to the present invention.
8 is a block diagram of an object area inspection unit according to the present invention.
9 to 11 are specific implementation views of the first inspector inspection module according to the present invention.
12 and 13 are specific implementation views of the second inspector inspection module according to the present invention.
14 is a configuration diagram of a data management unit according to the present invention.
15 is a configuration diagram of an image data management unit according to the present invention.
16 is a detailed implementation diagram of a data upload module according to the present invention.
17 is a detailed implementation diagram of a data allocation module according to the present invention.
18 is a block diagram of an item management unit according to the present invention.
19 is a detailed implementation diagram of the type management module according to the present invention.
20 is a detailed implementation diagram of an object management module according to the present invention.
21 is a detailed implementation diagram of the meta management module according to the present invention.
22 is a detailed implementation diagram of a user management unit according to the present invention.
23 is a block diagram of an authority management unit according to the present invention.
24 is a detailed implementation diagram of the authority registration module according to the present invention.
25 is a detailed implementation diagram of the authorization module according to the present invention.
26 is a detailed implementation diagram of a learning data detection unit according to the present invention.

Hereinafter, exemplary embodiments of the present invention will be described in detail with reference to the accompanying drawings. First of all, it should be noted that the same components or parts in the drawings represent the same reference numerals as possible. In describing the present invention, detailed descriptions of related known functions or configurations are omitted in order not to obscure the subject matter of the present invention.

1 is a block diagram of a system for providing learning data sets based on artificial intelligence according to an embodiment of the present invention.

An artificial intelligence-based learning data set providing system according to the present invention includes a user terminal 100, a manager terminal 200, and a learning data providing server 300, as shown in FIG. 1.

The user terminal 100 is a terminal capable of accessing a learning data providing server and using a tool provided by a corresponding learning data providing server according to the user's authority, and may be formed of a computer such as a desktop or a laptop or a smartphone.

When the user terminal 100 is made of a computer, the tool provided by the learning data providing server can be executed through the web running on the computer, and if the user terminal 100 is made of a smartphone, the corresponding smartphone The tool provided by the learning data providing server can be executed through the web running on the desktop or an application installed on the smartphone.

Specifically, the user terminal 100 may include a worker terminal 110, an inspector terminal 120, and a searcher terminal 130.

The worker terminal 110 is a terminal used by a worker among users, and may access a learning data providing server according to an input of a worker to designate an area for each object included in the image data.

The inspector terminal 120 may access the learning data providing server and inspect the object-specific area of the image data designated by the operator terminal by the operator's input. Such inspector terminal 120 includes a first inspector terminal 121a and It may include a second inspector terminal (121b).

The first inspector terminal 121a inspects the object-specific area of the image data designated by the worker terminal according to the operator's input, and if the object-specific area is incorrectly designated, it can be rejected to the operator terminal, and the corresponding object-specific area is correctly designated. In this case, the area for each object can be approved.

In addition, the second inspector terminal 121b inspects the object-specific area of the image data approved by the first inspector terminal according to the input of the operator, and if the inspection by the first inspector terminal is incorrect, it can reject the operator terminal. 1 If the inspection of the 1 inspector terminal is correct, the corresponding object-specific area is finally approved, so that the corresponding image data can be generated as learning image data to be learned by the artificial intelligence, as described later.

The searcher terminal 130 may access the learning data providing server, search for learning image data to be provided to artificial intelligence through a search tool provided by the learning data providing server, and then download and receive the searched learning image data. .

The manager terminal 200 connects to the learning data providing server and uses a tool provided by the learning data providing server to upload the image data, the item of the image data, the user of the user terminal, that is, an operator, an inspector (the first inspector, The second inspector), a searcher, and a terminal that manages the authority of the user, and may be formed of a computer such as a desktop, a notebook, or a smartphone, similar to the user terminal 100.

When the manager terminal 200 is made of a computer, it is possible to execute a tool provided by the learning data providing server through a web running on the computer, and when the manager terminal 200 is made of a smartphone, the corresponding smartphone The tool provided by the learning data providing server can be executed through the web running on the desktop or an application installed on the smartphone.

The learning data providing server 300 is connected to each other through a wired/wireless communication network 400 with the user terminal and the manager terminal to collect image data, that is, original image data, on a web page, and then for each object included in the image data. By designating an area, it is possible to generate training image data in which each object is divided.

2 is a block diagram of a learning data providing server according to the present invention.

Specifically, the learning data providing server 300 includes a member registration unit 310, a learning data generation unit 320, a data management unit 330, and a learning data detection unit 340, as shown in FIG. 2.

The member registration unit 310 may register member information of a worker, an inspector, and a searcher, respectively, to authenticate members.

3 is a configuration diagram of a member registration unit according to the present invention.

As shown in FIG. 3, the member registration unit 310 includes a worker member management unit 311 that registers member information of a worker to authenticate a member, an examiner member management unit 312 that registers member information of the examiner to authenticate a member, and searches It includes a searcher member management unit 313 for registering a member's member information to authenticate a member.

Specifically, the worker member management unit 311 may include a worker member registration module 311a, a worker member authentication module 311b, and a worker access permission module 311c.

The worker member registration module 311a may register member information of a worker.

The worker member registration module 311a accesses the learning data providing server 300 and receives personal information and other information of the worker who intends to use the tool provided by the object area designation unit 322 described later, and registers it as member information. can do.

The worker member authentication module 311b may authenticate member information of a worker.

The worker member authentication module 311b is the member information in which the member information input by the worker is pre-registered when the worker connects to the learning data providing server 300 and intends to use the tool provided by the object area designation unit 322 By performing the authentication process according to whether or not it is consistent with the operator can be connected to the learning data providing server 300.

The worker access permission module 311c may allow access to the learning data providing server to a worker who is member-authenticated among workers.

The worker access permission module 311c allows the worker to access the learning data providing server 300 when the worker trying to access the learning data providing server 300 is not authenticated as a member in the worker member authentication module 311b. Can block things.

On the other hand, when the worker is authenticated as a member in the worker member authentication module 311b, by allowing the worker to access the learning data providing server 300, the worker can use the tool provided by the object area designation unit 322 By using, an area for each object included in the image data can be designated.

The inspector member management unit 312 may include an inspector member registration module 312a, an inspector member authentication module 312b, and an inspector access permission module 312c.

The examinee member registration module 312a may register member information of the examinee.

The examiner member registration module 312a accesses the learning data providing server 300 to receive personal information and other information of the examiner who intends to use the tool provided by the object area inspection unit 323 to be described later, and register them as member information. can do.

The inspector member authentication module 312b may authenticate the inspector's member information.

When the examinee member authentication module 312b accesses the learning data providing server 300 and intends to use the tool provided by the object area inspection unit 323, the member information input by the examiner is pre-registered member information. By performing an authentication process according to whether or not it matches with, the examiner can be connected to the learning data providing server 300.

The inspector access permission module 312c may allow access to the learning data providing server to a member-authenticated inspector among inspectors.

The examiner access permission module 312c allows the examiner to access the learning data providing server 300 when the examiner attempting to access the learning data providing server 300 is not authenticated as a member in the examiner member authentication module 312b. Can block things.

On the other hand, when an inspector is authenticated as a member in the inspector member authentication module 312b, by allowing the inspector to access the learning data providing server 300, the corresponding worker can use the tool provided by the object area inspection unit 323. By using it, the area for each object included in the image data can be inspected.

The searcher member management unit 312 may include a searcher member registration module 313a, a searcher member authentication module 313b, and a searcher access permission module 313c.

The searcher member registration module 313a may register member information of a searcher.

The searcher member registration module 313a accesses the learning data providing server 300 to receive personal information and other information of the searcher who intends to use the tool provided by the learning data detection unit 340 to be described later, and register them as member information. can do.

The searcher member authentication module 313b may authenticate the searcher's member information.

When the searcher member authentication module 313b accesses the learning data providing server 300 and intends to use the tool provided by the learning data detection unit 340, the member information input by the searcher is already registered. The searcher can be connected to the learning data providing server 300 by performing an authentication process according to whether the information matches.

The searcher access permission module 313c may allow access to the learning data providing server to a searcher who is member-authenticated among searchers.

When the searcher accessing permission module 313c is not authenticated as a member in the searcher member authentication module 313b, the searcher attempting to access the learning data providing server 300 You can block access.

On the other hand, when the searcher is authenticated as a member in the searcher member authentication module 313b, by allowing the searcher to access the learning data providing server 300, the searcher can use the tool provided by the learning data detection unit 340. Using this, artificial intelligence can search and receive training image data to be learned.

After collecting the image data, the learning data generating unit 320 may generate training image data in which each object is classified as image data to be learned by artificial intelligence by designating an object-specific area included in the image data. .

4 is a block diagram of a learning data generator according to the present invention.

The learning data generation unit 320 includes an image data acquisition unit 321, an object area designation unit 322, an object area detection unit 323, and a training image data generation unit 324, as shown in FIG. Include.

The image data acquisition unit 321 may acquire image data for generating training image data by classifying only image data satisfying a specific object condition after collecting image data.

5 is a configuration diagram of an image data acquisition unit according to the present invention.

The image data acquisition unit 321 includes an image data collection module 321a, an image data classification module 321b, an image data storage module 321c, and a meta data generation module 321d, as shown in FIG. 5. do.

The image data collection module 321a may collect specific image data using a specific keyword on a web site.

Specifically, the image data collection module 321a may be composed of an engine that collects image data from a website, and this image data collection module 321a registers the URL and keyword of the website to be collected to collect image data from the corresponding site. Can be collected.

The image data classification module 321b may classify image data by filtering only image data that satisfies a specific object condition among image data collected by the image data collection module.

Specifically, the image data classification module 321b searches whether the collected image data is data corresponding to a specific object condition, such as a resolution, a photo with a person, a photo with only a face, and an animal, to classify the image data. I can.

The image data storage module 321c may temporarily store image data classified by the image data classification module.

The metadata generation module 321d may generate metadata of image data classified by an image data classification module.

Specifically, the metadata generating module 321d defines a data structure of image data, and may generate basic information such as size, latitude and longitude, resolution, and classification of the image.

The object area designation unit 322 may designate an area for each object included in the image data acquired by the image data acquisition unit.

Here, the object area designation unit 322 may display an image data list to designate an object area among image data on the operator terminal. In particular, when the operator terminal selects specific image data from the image data list, the object of the image data A tool to designate a star area can be provided.

6 and 7 are detailed implementation diagrams of an object area designating unit according to the present invention.

Specifically, when the worker terminal 110 accesses the learning data providing server, the object area designation unit 322 provides the worker terminal 110 with a list of image data that the worker should work with, as shown in FIG. If the operator selects, for example, a specific image ID or thumbnail image from the corresponding image list, as shown in FIG. 7, it can be moved to the object area designation page of the image, and the corresponding operator After designating an object area of the image using a tool provided on the page, it may be submitted to the object area inspection unit 323.

The object area inspection unit 323 may inspect an object-specific area of image data designated by the object area designation unit.

8 is a block diagram of an object area inspection unit according to the present invention.

The object area inspection unit 323 includes a first inspector inspection module 323a and a second inspector inspection module 323b, as shown in FIG. 8.

The first inspector inspection module 323a may inspect an object-specific area of image data designated by an operator terminal.

Here, the first inspector inspection module can display the worker list on the first inspector terminal, and when the first inspector terminal selects a specific worker from the worker list, it can display the work image list of the corresponding worker. , When the first inspector terminal selects a specific work image from the work image list, a tool for checking an object-specific area of the work image may be provided.

9 to 11 are specific implementation views of the first inspector inspection module according to the present invention.

Specifically, the first inspector inspection module 323a, when the first inspector terminal 121 accesses the learning data providing server, as shown in FIG. 9, the first inspector A list of workers to be inspected may be provided. When the first inspector selects, for example, a specific worker ID from the corresponding worker list, as shown in FIG. 10, a list of work images of the corresponding worker may be displayed.

Thereafter, when the first inspector selects, for example, a specific image ID or thumbnail image from the work image list, as shown in FIG. 11, the first inspector inspection module 323a is a page specifying the object area of the corresponding image. The first inspector can use the tool provided on the page to check the job details previously registered by the worker and then modify the job details.

In addition, the first inspector can request rework to the worker by stating the reason for rejection if the task content of the worker is the object of rejection, that is, if the object-specific area of the image is incorrectly designated. When the content is subject to approval, that is, when the object-specific area of the image is correctly designated, the object area designated in the image may be submitted to the second inspector inspection module 323b by approving the work.

The second inspector inspection module 323b may inspect an object-specific area of image data approved by the first inspector terminal.

Here, the second inspector inspection module 323b, when the second inspector terminal selects a specific worker from the worker list, can display a list of work images that have been inspected by the first inspector terminal of the corresponding worker. 2 When the inspector terminal selects a specific work image from the work image list, a tool for checking an object-specific area of the work image may be provided.

12 and 13 are specific implementation diagrams of a second inspector inspection module according to the present invention.

Specifically, when the second inspector inspection module 323b connects to the learning data providing server, the second inspector terminal 122 corresponds to the second inspector terminal 122 as shown in FIG. A list of workers to be inspected may be provided. When the second inspector selects, for example, a specific worker ID from the corresponding worker list, as shown in FIG. 13, a list of work images of the corresponding worker may be displayed.

Thereafter, when the second inspector selects, for example, a specific image ID or thumbnail image from the work image list, as shown in FIG. 11, the second inspector inspection module 323b is directed to the object area designation page of the corresponding image. It can be moved, but the second inspector can modify the work content after checking the work content previously registered by the worker using the tool provided on the page.

In addition, if the content of the worker's work is the object of rejection, the second inspector can request rework to the worker by specifying the reason for rejection, and on the contrary, when the content of the worker's work is subject to approval, the second inspector can finally approve the work. have.

The training image data generation unit 324 may generate the image data as training image data when the object area inspection unit approves the area for each object inspected.

Specifically, when the second inspector terminal checks an object-specific area of a specific work image of a specific worker in the worker list and finally approves the corresponding object-specific area, the learning image data generation unit 324 is based on the point coordinates of the corresponding area. The training image data, which is a mask file of, can be created.

The data management unit 330 may manage image data, an item of image data, a user of a user terminal, and a user's authority, respectively.

14 is a configuration diagram of a data management unit according to the present invention.

As shown in FIG. 14, the data management unit 330 includes an image data management unit 331, an item management unit 332, a user management unit 333 and an authority management unit 334.

The image data management unit 331 may register a folder in which image data is to be finally stored, upload and store image data in the folder, and allocate image data to a worker terminal to designate an object-specific area of the image data. .

15 is a configuration diagram of an image data management unit according to the present invention.

The image data management unit 331 includes a data upload module 331a and a data allocation module 331b, as shown in FIG. 15.

The data upload module 331a may register a folder and upload image data to the folder.

16 is a detailed implementation diagram of a data upload module according to the present invention.

Specifically, as illustrated in FIG. 16, the data upload module 331a may provide a tool for registering a folder in which image data is to be uploaded and for uploading specific image data by selecting a specific folder.

The data allocation module 331b may allocate specific image data uploaded to a folder to a specific worker terminal.

17 is a detailed implementation diagram of a data allocation module according to the present invention.

Specifically, as illustrated in FIG. 17, the data allocation module 331b may provide a tool for allocating specific image data to a specific worker by determining the number of allocated images after searching for an allocated image.

The item management unit 332 may manage an image type, which is an item of image data, an object number and a meta number corresponding to the image type, respectively.

18 is a block diagram of an item management unit according to the present invention.

As shown in FIG. 18, the item management unit 332 includes a type management module 332a, an object management module 332b, and a meta management module 332c.

The type management module 332a may register and modify an image type of image data.

19 is a detailed implementation diagram of the type management module according to the present invention.

Specifically, as shown in FIG. 19, the type management module 332a designates A, B, C, D, etc. as the type name of the image data, and the image type is a person, an upper body, and An image type can be registered by specifying a face, etc., or a corresponding image type can be modified or deleted.

The object management module 332b may register and modify the number of objects of an image type.

20 is a detailed implementation diagram of an object management module according to the present invention.

Specifically, the object management module 332b designates the foreground (all), the body (person), the upper body (upper body), the head (head), etc. as the object name (object), as shown in FIG. You can register the number of objects or modify or delete the number of objects.

The meta management module 332c may register and modify the meta number of an image type.

21 is a detailed implementation diagram of the meta management module according to the present invention.

Specifically, as shown in FIG. 21, the meta management module 332c designates person, location, type, etc. as meta names, and designates answer type, long sentence type, single selection type, multiple selection type, etc. as meta data types. You can register the meta number of an image type, or modify or delete the meta number.

The user management unit 333 may register and delete a corresponding user based on user information of a user accessing the learning data providing server, or modify and manage user information of a previously registered user.

22 is a detailed implementation diagram of a user management unit according to the present invention.

Specifically, the user management unit 333 provides a list of users for user ID, name, affiliation, and authority, as shown in FIG. 22, exposes a user registration pop-up when registering a user, and exposes a user deletion pop-up when deleting a user. I can make it.

The authority management unit 334 may register the user's authority, classify the user into an operator, an inspector, and a searcher, and then grant matching authority to each user.

23 is a block diagram of an authority management unit according to the present invention.

The authority management unit 334 includes an authority registration module 334a and an authority grant module 334b, as shown in FIG. 23.

The authority registration module 334a may register the user's authority and classify the user into an operator, an inspector, and a searcher.

24 is a detailed implementation diagram of the authority registration module according to the present invention.

Specifically, the permission registration module 334a provides a permission list for permission names and permission descriptions, as shown in FIG. 24, exposes permission registration pop-ups when registering permission for a specific user, and permission deletion pop-up when permission is deleted. Can be exposed.

The authority granting module 334b may grant each authority to an operator, an inspector, and a searcher.

25 is a detailed implementation diagram of the authorization module according to the present invention.

Specifically, as shown in FIG. 25, the authorization module 334b selects the authorization to be granted to the user, and then grants matching authorizations to the operator, the inspector, and the searcher, so that the operator ), the inspector makes the tool provided by the object area inspection unit 323 available, and the searcher makes the tool provided by the learning data detector 340 available.

The learning data detection unit 340 may provide a learning image data search tool to provide learning image data for artificial intelligence to learn according to a search result.

Here, the learning data detection unit 340 may provide a search tool capable of searching for learning image data based on an image type of the learning image data, an area for each object, and metadata, wherein the metadata is an expression of an object. , Gender and age may be included.

26 is a detailed implementation diagram of a learning data detection unit according to the present invention.

Specifically, when the searcher terminal 130 connects to the learning data providing server, the learning data detection unit 340, as shown in FIG. 26, provides an image type, an object-specific area, and metadata (expression, gender, age, etc.). By providing a search tool including ), the searcher can search for and receive training image data for artificial intelligence to learn through the search tool.

As described above, with reference to the drawings illustrating an artificial intelligence-based learning data set providing system according to the present invention, the present invention is not limited by the embodiments and drawings disclosed in the present specification, and the technical idea of the present invention It goes without saying that various modifications can be made by those skilled in the art within the range.

100: user terminal 110: worker terminal
120: inspector terminal 121: first inspector terminal
122: second inspector terminal 130: searcher terminal
200: administrator terminal 300: learning data providing server
310: Member registration department 311: Worker member management department
311a: Worker member registration module 311b: Worker member authentication module
311c: Worker access permission module 312: Inspector member management unit
312a: Examiner member registration module 312b: Examiner member authentication module
312c: Inspector access permission module 313: Searcher member management unit
313a: searcher member registration module 313b: searcher member authentication module
313c: searcher access permission module 320: learning data generation unit
321: image data acquisition unit 321a: image data acquisition module
321b: image data classification module 321c: image data storage module
321d: metadata generation module 322: object area designation unit
323: object area inspection unit 323a: first inspector inspection module
323b: second inspector inspection module 324: learning data generation unit
330: data management unit 331: image data management unit
331a: data upload module 331b: data allocation module
332: item management unit 332a: type management module
332b: object management module 332c: meta management module
333: user management unit 334: authority management unit
334a: permission registration module 334b: permission grant module
340: learning data detection unit 400: wired/wireless communication network

Claims (15)

User terminal;
An administrator terminal for managing the user of the user terminal and the authority of the user; And
A learning data providing server that is connected to each other through a wired/wireless communication network with the user terminal and the manager terminal, collects image data and designates a region for each object included in the image data to generate learning image data in which each object is classified; Including,
The user terminal,
An operator terminal accessing the learning data providing server and designating an object-specific area included in the image data according to an input of an operator;
An inspector terminal for accessing the learning data providing server and inspecting an object-specific area of the image data designated by the operator terminal according to an input of an operator; And
Includes; a searcher terminal connected to the learning data providing server to search for and receive learning image data to be provided to artificial intelligence through a search tool,
The inspector terminal may include: a first inspector terminal for generating approval information for inspecting and approving an object-specific area of the image data designated by the operator terminal according to an input of the operator, or generating rejection information for rejecting and transmitting it to the operator terminal; And a second in which the first inspector terminal inspects the object-specific area of the image data for which the approval information was generated by the input of the worker, and generates final approval information for final approval, or generates rejection information for rejection and delivers it to the worker terminal. Including; inspector terminal;
The learning data providing server,
A member registration unit for registering member information of a worker, an inspector, and a searcher, respectively, to authenticate a member;
After collecting the image data, a learning data generation unit that designates an area for each object included in the corresponding image data to generate training image data in which each object is classified as image data to be learned by the artificial intelligence;
A data management unit that manages the image data, items of image data, a user of a user terminal, and a user's authority, respectively; And
Including; a learning data detection unit that provides a search tool for the learning image data to provide learning image data to be learned by the artificial intelligence according to the search result.
The learning data generation unit,
An image data list to designate an area for each object among the image data is displayed on the worker terminal, and when the worker terminal selects specific image data from the image data list, a tool for designating an object-specific area of the image data is provided,
When the operator terminal displays an image data list in which an object-specific area is specified on the inspector terminal, and the inspector terminal selects specific image data from the image data list, it provides a tool to check the object-specific area of the image data. If approved, an artificial intelligence-based learning data set providing system, characterized in that the image data is generated as training image data.
delete The method of claim 1,
The member registration unit,
A worker member management unit that registers member information of the worker to authenticate a member;
An inspector member management unit for registering member information of the inspector to authenticate a member; And
Including; a searcher member management unit for registering the member information of the searcher to authenticate a member,
The worker member management unit,
A worker member registration module for registering member information of the worker;
A worker member authentication module for authenticating member information of the worker; And
Includes; a worker access permission module for allowing access to the learning data providing server to a member of the worker authenticated,
The inspector member management unit,
An inspector member registration module for registering member information of the inspector;
An inspector member authentication module for authenticating the inspector's member information; And
Including; a reviewer access permission module for allowing access to the learning data providing server to a member-authenticated reviewer among the reviewers,
The searcher member management unit,
A searcher member registration module for registering member information of the searcher;
A searcher member authentication module for authenticating member information of the searcher; And
And a searcher access permission module for allowing access to the learning data providing server to a searcher who has been authenticated as a member of the searchers.
The method of claim 1,
The learning data generation unit,
An image data acquisition unit that collects image data and then classifies and acquires only image data satisfying a specific object condition;
An object area designation unit that designates an area for each object included in the image data acquired by the image data acquisition unit;
An object area inspection unit that inspects an object-specific area of the image data designated by the object area designation unit; And
And a learning image data generation unit that generates corresponding image data as training image data when the object area inspection unit approves the area for each object. The system for providing a learning data set based on artificial intelligence, comprising: a.
The method of claim 4,
The image data acquisition unit,
An image data collection module for collecting specific image data by using a specific keyword on a website;
An image data classification module for classifying image data by filtering only image data satisfying a specific object condition among image data collected by the image data collection module;
An image data storage module temporarily storing the image data classified by the image data classification module; And
And a meta data generation module that generates meta data of image data classified by the image data classification module.
The method of claim 5,
The object area designation unit,
Displaying an image data list to designate an object area among the image data on the operator terminal,
When the operator terminal selects specific image data from the image data list, an artificial intelligence-based learning data set providing system, characterized in that it provides a tool for designating an object-specific area of the image data.
The method of claim 6,
The object area inspection unit,
A first inspector inspection module for inspecting an object-specific area of the image data designated by the operator terminal; And
And a second inspector inspection module for inspecting an object-specific area of the image data approved by the first inspector terminal.
The method of claim 7,
The first inspector inspection module,
Display the worker list on the first inspector terminal,
When the first inspector terminal selects a specific worker from the worker list, displays a list of work images of the worker,
When the first inspector terminal selects a specific work image from the work image list, an artificial intelligence-based learning data set providing system, characterized in that it provides a tool for checking an object-specific area of the work image.
The method of claim 8,
The second inspector inspection module,
Display the worker list on the second inspector terminal,
When the second inspector terminal selects a specific worker from the worker list, a list of work images that have been inspected by the first inspector terminal of the worker are displayed, and
When the second inspector terminal selects a specific work image from the work image list, an artificial intelligence-based learning data set providing system, characterized in that it provides a tool for checking an object-specific area of the work image.
The method of claim 9,
The training image data generation unit,
When the second inspector terminal checks the area for each object of the specific work image of a specific worker in the worker list and finally approves the area for each object,
An artificial intelligence-based training data set providing system, characterized in that generating training image data that is a mask file based on point coordinates of a corresponding area.
The method of claim 10,
The data management unit,
An image data management unit that registers a folder in which the image data will be finally stored, uploads and stores image data in the folder, and allocates image data to a worker terminal to designate an object-specific area of the image data;
An item management unit for managing an image type, which is an item of the image data, an object number and a meta number corresponding to the image type, respectively;
A user management unit that registers and deletes a corresponding user based on user information of a user accessing the learning data providing server, or modifies and manages user information of a previously registered user; And
And an authority management unit that registers the user's authority to classify the user into an operator, an inspector, and a searcher, and then grants matching authority to each user.
The method of claim 11,
The image data management unit,
A data upload module for registering the folder and uploading image data to the folder; And
And a data allocation module for allocating specific image data uploaded to the folder to a specific worker terminal.
The method of claim 12,
The item management unit,
A type management module for registering and modifying an image type of the image data;
An object management module for registering and modifying the number of objects of the image type; And
And a meta management module for registering and modifying the meta number of the image type.
The method of claim 13,
The authority management unit,
An authority registration module for registering the user's authority and classifying the user into an operator, an inspector, and a searcher; And
And an authorization module for granting respective authorizations to the operator, the inspector, and the searcher. 14. A system for providing learning data sets based on artificial intelligence, comprising: a.
The method of claim 14,
The learning data detection unit,
Providing a search tool capable of searching for learning image data based on an image type of the learning image data, an area for each object, and metadata,
The meta data is an artificial intelligence-based learning data set providing system, characterized in that the expression, gender, and age of the object.

Images