KR20230068704A - Learning data integrated management platform system based on artificial intelligence - Google Patents

Abstract

The present invention relates to an artificial intelligence-based learning data integration management platform device, which collects image-based original data related to learning data from an unspecified number of data providers through the cloud, and transfers the collected image-based original data to an unspecified number of unspecified data providers. It relates to an artificial intelligence-based learning data integration management platform device that provides a platform to perform annotation work from a data processor. For each project created for this purpose, it provides a collection platform module that collects image-based original data that requires artificial intelligence learning from different data providers through a network, and an original source that stores multiple image-based original data collected. A data collection unit, an original data filtering unit that compares the image-based original data provided from the collection platform module with a predetermined image condition and filters the original data based on the filtered image, and adjusts the size, capacity, and format of the original data based on the predetermined format. Original data refinement unit that provides a refinement platform module that manages refinement, labeling or tagging the refined image-based original data in the refinement platform module, and setting the annotation area required for artificial intelligence learning to the refined image-based original data The original data processing unit that provides the annotation platform module, and the learning data generation platform module that receives the original data based on the processed image provided by the annotation platform module, and the processed image depending on the performance and quality of artificial intelligence machine learning Provides an artificial intelligence machine learning platform module that performs artificial intelligence machine learning based on the learning data provided by the learning data generation unit and learning data generation platform module that newly input the original data, and improves the performance of artificial intelligence machine learning. Provides a quality control platform module that manages the quality of learning data according to the measured artificial intelligence machine learning unit, the distribution of processed image-based original data input to the learning data generation platform module, and the performance measurement of the artificial intelligence machine learning platform module and a learning data quality management unit that increases or decreases the amount of processed image-based original data input to the artificial intelligence machine learning platform module according to quality control. An artificial intelligence-based integrated learning data management platform. device is started.

Description

Learning data integrated management platform system based on artificial intelligence}

The present invention relates to an artificial intelligence-based learning data integration management platform device, and more particularly, to collect image-based original data related to learning data from a plurality of unspecified data providers through the cloud, and collect image-based original data It relates to an artificial intelligence-based learning data integrated management platform device that provides a platform to perform annotation work from an unspecified number of data processors.

In general, acquiring various original data for machine learning and processing the acquired original data is a cumbersome task that requires a lot of time and manual work. In addition, when the processed original data is used for machine learning, input of additional original data may be required to improve the performance scale and quality of machine learning. There is a demand for a platform capable of acquiring various original data for machine learning from the accessed and unspecified majority, providing a platform to process the original data, and performing machine learning under the platform.

KR 10-2179586 KR 10-2232880 KR 10-2108400

Therefore, the present invention was created to solve the above problems, and provides a platform to receive various image-based original data required for machine learning from a number of unspecified data providers and process the supplied original data, Its purpose is to provide a platform that allows machine learning to be integrated through processed original data.

In addition, an object of the present invention is to provide an integrated platform for repeatedly reacquiring, processing, and managing machine learning in order to solve the hassle of supplementing the learning data again when the performance scale is degraded through machine learning. There is a purpose.

However, the objects of the present invention are not limited to the above-mentioned objects, and other objects not mentioned will be clearly understood by those skilled in the art from the description below.

An object of the present invention described above is to provide a collection platform module that collects image-based original data that requires artificial intelligence learning for each project created from different data providers through a network, and An original data collection unit that stores original data, an original data filtering unit that compares and filters the image-based original data provided by the collection platform module with preset image conditions, and the size, capacity, and format of the original data based on the filtered image. Original data refinement unit that provides a refinement platform module that refines and manages the original data according to a predetermined format, labels or tags the refined image-based original data in the refinement platform module, and divides the annotation area necessary for artificial intelligence learning into a refined image-based Original data processing unit that provides an annotation platform module that sets the area in the original data, and a learning data generation platform module that receives original data based on processed images provided by the annotation platform module, providing performance and quality of artificial intelligence machine learning Provides an artificial intelligence machine learning platform module that performs artificial intelligence machine learning based on the learning data provided by the learning data generation unit and learning data generation platform module that receives new input of processed image-based original data, The artificial intelligence machine learning unit that measures the performance of intelligent machine learning, the distribution of processed image-based original data input to the learning data generation platform module, and the quality of the learning data are managed according to the performance measurement of the artificial intelligence machine learning platform module. Provides a quality control platform module and includes a learning data quality management unit that increases or decreases the amount of processed image-based original data input to the artificial intelligence machine learning platform module according to quality control. This can be achieved by providing an integrated learning data management platform device.

In addition, the collection platform module, original data filtering unit, refinement platform module, and annotation platform module are sequentially connected to each other through a pipeline, so that all image-based original data stored in lower-level modules are sequentially batched at the request of the upper-level module. so that it can be connected.

In addition, the original data filtering unit includes a primary determination unit having a primary filtering module that primarily determines whether the image-based original data transmitted from the collection platform module meets a predetermined primary image condition, and an artificial intelligence machine learning platform. It has an artificial intelligence machine learning module built through the module, and includes an artificial intelligence model determination unit that secondarily determines whether the image-based original data that has passed through the primary filtering module meets a predetermined secondary image condition.

In addition, the original data processing unit provides an editing platform module equipped with an editing tool to set an object area required for learning in one area of the image-based original data provided by the refinement platform module, and includes metadata of the image-based original data. An annotation processing unit that provides an input platform module equipped with an input tool to inspect the image-based original data processed by the annotation processing unit according to preset inspection conditions, and image-based original data that satisfies the preset conditions is learned It is transmitted to the data generation platform module, and the image-based original data that does not satisfy the conditions is retransmitted to the annotation processing unit to include an annotation verification unit to newly set the object area.

In addition, the artificial intelligence machine learning unit has an artificial intelligence machine learning platform module that performs artificial intelligence machine learning based on the learning data provided by the learning data generation platform module, and the performance of artificial intelligence machine learning. It measures according to the set scale, and includes an artificial intelligence model performance measurement unit that adjusts the ratio of the amount of training data set and the amount of verification data set transmitted from the learning data generation platform module according to the performance measurement value.

According to the present invention as described above, various image-based original data required for machine learning are supplied from a plurality of unspecified data providers, a platform is provided to process the supplied original data, and machine learning is integrated through the processed original data. There is an effect that can build a platform that can be achieved.

The following drawings attached to this specification illustrate a preferred embodiment of the present invention, and together with the detailed description of the present invention serve to further understand the technical idea of the present invention, the present invention is limited to those described in the drawings. It should not be construed as limiting.
1 is a diagram schematically showing the overall configuration of an artificial intelligence-based integrated learning data management platform device according to an embodiment of the present invention;
2 is a diagram showing an original data filtering unit according to an embodiment of the present invention;
3 is a diagram showing an original data processing unit according to an embodiment of the present invention;
4 is a diagram showing setting an object area in image-based original data and inputting metadata according to an embodiment of the present invention;
5 is a diagram showing that a pipeline is connected between an upper-level module and a lower-level module according to an embodiment of the present invention so that original data of a lower-level module can be automatically uploaded to an upper-level module;
6 is a diagram illustrating a sequential platform flow of an artificial intelligence-based integrated learning data management platform device according to an embodiment of the present invention.

Hereinafter, a preferred embodiment of the present invention will be described with reference to the drawings. In addition, the embodiment described below does not unduly limit the contents of the present invention described in the claims, and the entire configuration described in the present embodiment cannot be said to be essential as a solution to the present invention. In addition, descriptions of matters obvious to prior art and those skilled in the art may be omitted, and descriptions of these omitted components (methods) and functions may be sufficiently referred to within the scope not departing from the technical spirit of the present invention.

An artificial intelligence-based integrated learning data management platform device according to an embodiment of the present invention relates to a device for providing a platform capable of obtaining image-based learning data from an unspecified number of people and performing machine learning through a cloud or network. will be. Hereinafter, an artificial intelligence-based integrated learning data management platform device according to an embodiment of the present invention will be described in detail with reference to the accompanying drawings.

An artificial intelligence-based learning data integrated management platform device (hereinafter referred to as a platform device) according to an embodiment of the present invention may include a cloud server.

As shown in FIGS. 1 and 2, the artificial intelligence-based integrated learning data management platform device according to an embodiment of the present invention includes an original data collection unit 110, an original data filtering unit 115, and an original data refinement unit ( 120), an original data processing unit 130, a learning data generating unit 140, an artificial intelligence machine learning unit 150, and a learning data quality management unit 160.

The original data collection unit 110 according to an embodiment of the present invention includes a collection platform module. The collection platform module 110 may receive data input from a number of unspecified data providers through access to a network or Internet network. That is, an unspecified number of data providers can transmit image-based learning data to the collection platform module through their smartphones. As another example, a number of unspecified data providers may transmit image-based learning data by accessing the collection platform module through a cloud server.

The aggregation platform modules 110 may be categorized per project. That is, the first collection platform module collects the original data based on the first image, the second collection platform module collects the original data based on the second image, and the nth collection platform module collects the original data based on the n-th image. can be categorized. At this time, as an example, the first image-based original data may be data related to "learning data for diagnosing a disease of a dog", and the original data based on the second image may be data related to "learning data for diagnosing a disease of a chicken". can be data. However, the above-described example is only an example, and various image-based original data can be learned and managed by the platform device of the present invention.

An application (hereinafter referred to as an app) that can be linked with the collection platform module 110 of the present invention may be installed on a smartphone possessed by an unspecified plurality of data providers. A camera module, a communication module, and a control module may be provided in smartphones possessed by a number of unspecified data providers. Therefore, the data provider obtains image-based original data using the camera module, and collects it through the communication module after generating an evidence image that can prove the place or date of acquisition of the image-based original data through the control module. It is transmitted to the platform module 110.

As shown in FIG. 2 , the original data filtering unit 115 according to an embodiment of the present invention includes a primary determination unit and an artificial intelligence model determination unit.

The primary determination unit includes a primary filtering module, and the primary filtering module inspects the image-based original data provided from the collection platform module 110 and primarily filters it. That is, the primary filtering module provided in the primary determination unit firstly determines whether image-based original data provided from a data provider meets a predetermined primary image condition. In this case, the primary image condition is image-based original data and proof images provided from a plurality of unspecified data providers, and it may be analyzed to determine whether or not the primary image condition is satisfied. That is, the data provider should transmit image-based original data related to “A” to the “A collection platform module” of the original data collection unit, but may transmit image-based original data related to “B” by mistake. As an example, mistakes of such data providers are primarily filtered and inspected. At this time, the image-based original data passed through the primary filtering module is re-entered into the artificial intelligence machine learning module described later, and the image-based original data filtered by the primary filtering module is retransmitted to the original data collection unit. They are classified from each other by being retransmitted together with . The original data collection unit 110 discards the retransmitted image-based original data with reference to the reason for retransmission or moves it to a collection platform module suitable for itself.

The artificial intelligence model judgment unit has an artificial intelligence machine learning module. The artificial intelligence machine learning module is a machine learning module built through the artificial intelligence machine learning platform module 150. The artificial intelligence machine learning module compares the image-based original data judged and filtered by the primary judgment unit with a predetermined secondary image condition to secondarily inspect and filter. At this time, the secondary image condition is the original data based on the first-order filtered image in the primary filtering module and the evidence image (at this time, the evidence image is predefined in a format that is easy for the artificial intelligence machine learning module to analyze). It may be determined whether or not a set secondary image condition is met. The second inspection method may be the same as the above-described first inspection method of the filtering module.

The above-described primary determination unit and artificial intelligence model determination unit sequentially inspect and filter image-based original data provided by a plurality of unspecified data providers in the first and second stages. At this time, as an example, the first judgment unit may perform a first inspection on the connection error of the pipeline, and the artificial intelligence model judgment unit checks the pipeline connection error. It is possible to inspect and filter whether the intelligent machine learning module meets a predetermined image condition (predefined in a format that is easy to analyze).

Here, the connection error of the pipeline may be an error when the original data filtering unit 115, which is an upper module, calls the collected data to the original data collection unit of the related lower module. That is, as an example, if the original data of the "project related to disease A" is filtered in the original data filtering unit 115, which is an upper module, it is included in the "project related to disease B" of the original data collection unit 110, which is a lower module. It may be a case of an error in which the original data that has been processed is connected to the pipeline.

Meanwhile, the image-based original data secondarily passed by the artificial intelligence machine learning module is input to the original data refining unit 120 to be described later, and the image-based original data filtered by the artificial intelligence machine learning module is returned again. By being retransmitted together with the reason for retransmission to the original data collection unit, they are classified. The original data collection unit 110 discards the retransmitted image-based original data with reference to the reason for retransmission or moves it to a collection platform module suitable for itself.

The original data purification unit 120 according to an embodiment of the present invention includes a purification platform module. The refinement platform module 120 receives image-based original data filtered secondarily by the artificial intelligence machine learning module.

The refinement platform module 120 formats the received image-based original data to conform to the format of the preset learning data. That is, the size, capacity, and data format of the transmitted original data are standardized. The reason for standardizing the format in this way is that various types of data transmitted from a number of unspecified data providers must be standardized to meet predetermined standards so that they can be used as learning data, and furthermore, annotation of image-based original data is possible. am.

The original data processing unit 130 according to an embodiment of the present invention includes an annotation platform module. The annotation platform module 130 labels or tags the image-based original data refined in the refinement platform module 120, and sets an annotation region necessary for artificial intelligence learning to the refined image-based original data. To this end, the original data processing unit 130 includes an annotation processing unit and an annotation inspection unit.

The annotation processing unit includes an editing platform module equipped with an editing tool to set an object area required for learning in one area of the image-based original data provided by the refinement platform module 120 and an input tool including metadata of the image-based original data. This equipped input platform module is provided. The editing platform module provides a bounding box or polygon segment to set an object region in image-based original data. In addition, the input platform module provides an input tool so that metadata of image-based original data can be input, as shown in FIG. 4 . Metadata is labeling of image-based original data, and may include, for example, a disease area, a photographing posture, a disease name, and the like.

As an example, as shown in FIG. 4 , the object area 131a required for learning is set in one area of the image-based original data 131 by utilizing a bounding box or a polygon segment. The set object area is used for machine learning in the artificial intelligence machine learning unit 150 to be described later.

On the other hand, the annotation verification unit inspects the image-based original data processed by the annotation processing unit according to the preset verification conditions, and transmits the image-based original data that satisfies the preset conditions to the learning data generation platform module 140, Image-based original data that does not satisfy the conditions is retransmitted to the annotation processing unit so that a new object area is set.

At this time, the annotation verification unit may determine whether or not the preset conditions are satisfied according to the following conditions. That is, the annotation verification unit may make the determination by comparing the content included in the aforementioned metadata with the object area set by the bounding box or polygon segment. To this end, the annotation verification unit may include an artificial intelligence machine learning module. The artificial intelligence machine learning module may determine whether the setting of the object area is correct or the description of the metadata is correct by analyzing the image set as the object area and comparing it with the contents described in the metadata. If there is an error in any one of the description of the object area or metadata, it can be corrected by transmitting related data to the annotation processing unit again, including an error comment.

On the other hand, as shown in Figure 4, the annotation processing unit can be connected and connected to the smart phones of a plurality of unspecified data processors. Alternatively, an unspecified number of data processors may directly connect to the annotation processing unit through the web. Therefore, an unspecified number of data processors are connected to the annotation processing unit through the Internet or network and directly set an object area of image-based original data stored in the annotation processing unit. To this end, an app may be installed in a smartphone possessed by a plurality of unspecified data processors, and the annotation processing unit may be directly connected through the installed app.

As shown in FIG. 1, the learning data generating unit 140 includes a learning data generating platform module. The learning data generation platform module 140 receives the processed image-based original data provided by the annotation platform module 130, and adds new processed image-based original data according to the performance and quality of artificial intelligence machine learning. receive input That is, the learning data generation platform module additionally receives original data based on the processed image according to the quality control of the learning data quality control unit 160 . By receiving new image-based original data, the performance of artificial intelligence machine learning can be improved, and the quality of learning can also be improved accordingly.

The artificial intelligence machine learning unit 150 according to an embodiment of the present invention shown in FIG. 1 includes an artificial intelligence machine learning platform module. The artificial intelligence machine learning platform module 150 performs artificial intelligence machine learning based on the image-based learning data including the object area and metadata provided by the learning data generation platform module 140, and performs artificial intelligence machine learning. measure performance To this end, the artificial intelligence machine learning unit 150 includes an artificial intelligence model learning unit and an artificial intelligence model performance measuring unit.

The artificial intelligence model learning unit includes an artificial intelligence machine learning platform module that performs artificial intelligence machine learning based on the learning data provided by the learning data generation platform module 140. An artificial intelligence machine learning module is built according to the machine learning of the artificial intelligence machine learning platform module. The constructed artificial intelligence machine learning module can be used in the artificial intelligence model judgment unit and annotation inspection unit as described above.

The AI model performance measurement unit measures the performance of artificial intelligence machine learning. Measures of performance are based on accuracy, precision and recall. A measure of performance measurement may be referenced within a range that does not deviate from the technical spirit of the present invention.

The ratio between the amount of the training data set transmitted from the learning data generation platform module 140 and the amount of the verification data set is adjusted according to whether the performance of the artificial intelligence model performance measuring unit is measured. That is, as an example, when the ratio of the learning data set and the verification data set of the image-based original data input from the learning data generation platform module 140 is set to a ratio of 7:3 by the artificial intelligence machine learning platform module 150 You can adjust these ratios in . In addition, as will be described later, if the performance measurement of artificial intelligence machine learning is poor, quality control may also be poor, and therefore, the amount of image-based original data input to the learning data generator 140 as a whole can be increased.

As shown in FIG. 1, the learning data quality management unit 160 according to an embodiment of the present invention includes a quality management platform module. The quality control platform module 160 manages the quality of the learning data according to the distribution of the processed image-based original data input to the learning data generation platform module 140 and the performance measurement of the artificial intelligence machine learning platform module 150, , Increase or decrease the amount of processed image-based original data input to the artificial intelligence machine learning platform module 150 according to quality control.

At this time, in order to increase the amount of image-based original data input to the artificial intelligence machine learning platform module 150, the amount of original data collected by the original data collection unit 110 must be increased. On the other hand, if image-based original data itself is rare data that is difficult to obtain, it is difficult to collect data from a data provider, so the amount of original data can be increased through a data augmentation technique for machine learning.

The quality management platform module 160 tracks and manages the distribution of image-based original data. In other words, by tracking the distribution of image-based original data where machine learning is well learned and image-based original data where machine learning is not well performed, the amount of image-based original data where machine learning is not well performed can be further increased. can be controlled.

On the other hand, as shown in FIG. 5, as an example, the original data collection unit 110, the original data purification unit 120, and the original data processing unit 130 may be sequentially called a lower stage module and an upper stage module. Since the module in the upper stage and the module in the upper stage are connected to each other through a pipeline, the data of the immediately preceding lower stage module can be automatically uploaded to the upper stage module. As an example, as shown in FIG. 5 , the original data collection unit 110 collects and stores image-based original data corresponding to various projects from data providers. As an example of these, when image-based original data for recognizing “A” is collected as much as {1,...a}, refinement for recognizing “A” in the original data refining unit 120, which is an upper module. When a project is started, the original data of {1,...a} is automatically uploaded to the upper module according to the pipeline connection, and based on the original data of {1,...a}, {1,...a} is automatically uploaded. M}'s refined image-based original data is created. The pipeline connection from the original data refining unit 120 to the original data processing unit 130 can be explained using the same principle, and although not shown in FIG. 5, the same principle can be applied to the original data filtering unit 115. .

In describing the present invention, descriptions of matters obvious to those skilled in the art and those skilled in the art may be omitted, and descriptions of these omitted components (methods) and functions will be sufficiently referred to within the scope that does not depart from the technical spirit of the present invention. You will be able to. In addition, the above-described components of the present invention have been described for convenience of description of the present invention, but components not described herein may be added within a range that does not deviate from the technical spirit of the present invention.

The description of the configuration and function of each part described above has been separately described for convenience of explanation, but any one configuration and function may be implemented by integrating into other components or implemented in more subdivided form, if necessary.

Although the above has been described with reference to one embodiment of the present invention, the present invention is not limited thereto, and various modifications and applications are possible. That is, those skilled in the art will easily understand that many modifications are possible without departing from the gist of the present invention. In addition, when it is determined that the detailed description of known functions and their configurations related to the present invention or the coupling relationship of each component of the present invention may unnecessarily obscure the gist of the present invention, it should be noted that the detailed description is omitted. something to do.

100: artificial intelligence learning data integrated management platform unit
110: original data collection unit
115: original data filtering unit
120: original data purification unit
130: original data processing unit
131: image-based original data
131a: object area
140: learning data generation unit
150: artificial intelligence machine learning unit
160: learning data quality control unit

Claims (5)

For each project created, it provides a collection platform module that collects image-based original data that requires AI learning from different data providers through a network, and collects original data that stores multiple image-based original data collected. wealth,
an original data filtering unit for filtering by comparing the image-based original data provided by the collection platform module with a preset image condition;
Original data refining unit providing a refining platform module for refining and managing the size, capacity, and format of the filtered image-based original data according to a predetermined format;
An original data processing unit that provides an annotation platform module for labeling or tagging image-based original data refined in the refinement platform module and setting an annotation region necessary for artificial intelligence learning in the refined image-based original data;
Provides a learning data generation platform module that receives the processed image-based original data provided by the annotation platform module, and receives new input of the processed image-based original data according to the performance and quality of artificial intelligence machine learning learning data generation unit,
An artificial intelligence machine learning unit that provides an artificial intelligence machine learning platform module that performs artificial intelligence machine learning based on the learning data provided by the learning data generation platform module, and measures the performance of the artificial intelligence machine learning;
Provide a quality control platform module that manages the quality of learning data according to the distribution of processed image-based original data input to the learning data generation platform module and the performance measurement of the artificial intelligence machine learning platform module, and according to quality control An artificial intelligence-based learning data integrated management platform device comprising a learning data quality management unit that increases or decreases the amount of processed image-based original data input to the artificial intelligence machine learning platform module.
According to claim 1,
The collection platform module, the original data filtering unit, the refinement platform module, and the annotation platform module are sequentially connected to each other in a pipeline, so that all image-based original data stored in the lower-stage module are sequentially batched at the request of the upper-stage module. An artificial intelligence-based integrated learning data management platform device characterized in that it can be connected.
According to claim 1,
The original data filtering unit,
a primary determination unit having a primary filtering module that primarily determines whether or not the image-based original data transmitted from the collection platform module satisfies a predetermined primary image condition;
It has an artificial intelligence machine learning module built through the artificial intelligence machine learning platform module, and secondarily determines whether the image-based original data that has passed through the first filtering module meets a predetermined second image condition. An artificial intelligence-based learning data integrated management platform device comprising an intelligent model judgment unit.
According to claim 3,
The original data processing unit,
Provides an editing platform module equipped with an editing tool to set an object area necessary for learning in one area of the image-based original data provided by the refinement platform module, and an input tool including metadata of the image-based original data An annotation processing unit providing the provided input platform module;
The image-based original data processed by the annotation processing unit is inspected according to the preset inspection conditions, and the image-based original data that meets the preset conditions is transmitted to the learning data generation platform module, and the image-based data that does not satisfy the conditions The artificial intelligence-based learning data integrated management platform device, characterized in that it comprises an annotation verification unit for retransmitting the original data to the annotation processing unit so that a new object area is set.
According to claim 4,
The artificial intelligence machine learning unit,
An artificial intelligence model learning unit having an artificial intelligence machine learning platform module for performing artificial intelligence machine learning based on the learning data provided by the learning data generation platform module;
Artificial intelligence model performance that measures the performance of artificial intelligence machine learning according to a preset scale and adjusts the ratio of the amount of training data set and the amount of verification data set transmitted from the learning data generation platform module according to the performance measurement value Artificial intelligence-based integrated learning data management platform device comprising a measurement unit.

Images