Classification system for unstructured data based on ai

Abstract

An AI-based classification system and method for unstructured data are disclosed. An AI-based classification system for unstructured data according to one embodiment of the present invention comprises: a manual classification unit provided to a classification worker to manually classify sample images extracted from unstructured image data subject to classification work according to a preset classification class; an AI model training unit that automatically classifies the sample images classified by the manual classification unit according to the preset classification class through an AI model to train the AI ​​model; and an automatic classification unit that automatically classifies all images of the unstructured image data according to the preset classification class through the AI ​​model trained by the AI ​​model training unit.

Description

Technology Field

[0001] The present invention relates to an AI (Artificial Intelligence)-based classification system and method for image data, and more specifically, to an AI-based classification system and method for unstructured image data with ambiguous classification criteria. Background Technology

[0003] The research and development projects that supported this invention are as follows.

[0004] [Assignment No.] 2005221027

[0005] [Ministry Name] National Information & Communication Technology Promotion Agency (NIPA)

[0006] [Project Management Agency Name] Jeonju Information & Culture Industry Promotion Agency

[0007] [Research Project Name] Regional SW Service Commercialization Support Project

[0008] [Research Project Title] Development of an Automated System for Material Mix Quality Inspection at Smart Construction Sites Based on Video AI Analysis Technology

[0009] [Name of Project Performing Organization] Pointi Co., Ltd.

[0010] [Research Period] April 1, 2022 ~ December 31, 2023

[0011] In general, class classification of structured image data is performed using clear classification criteria, resulting in high accuracy and reliability of the classification results.

[0012] On the other hand, class classification of unstructured image data, for which the industrial demand has recently been increasing, faces significant difficulties in improving the accuracy and reliability of the results. This is because unstructured images lack consistent patterns, making classification criteria ambiguous, and individual differences occur among classifiers.

[0013] Accordingly, there is a need for research and development of AI-based classification systems and methods that apply AI models to unstructured image data to perform class classification while ensuring high accuracy and reliability in the classification results.

[0014] Related prior art documents include Registered Patent Publication No. 10-2391501 (Title of Invention: System and Method for Classifying Irregular Recyclables Using Deep Learning, Date of Publication: April 27, 2022) and Published Patent Publication No. 10-2022-0117039 (Title of Invention: Method for Manufacturing Waste Solidified Body and Deep Learning-Based Automatic Classification Method and System for Bridge Components, Date of Publication: August 23, 2022). The problem to be solved

[0016] The objective of the present invention is to provide an AI-based classification system and method for unstructured image data that can reduce the time required for classification work while ensuring high accuracy and reliability by applying an AI model when classifying unstructured image data with ambiguous classification criteria according to a target classification class. means of solving the problem

[0018] The above objective is achieved by an AI-based classification system for unstructured data, characterized in that, according to one embodiment of the present invention, it comprises a manual classification unit provided to a classification worker to manually classify sample images extracted from unstructured image data subject to classification work according to a preset classification class, and an AI model training unit that automatically classifies the sample images classified by the manual classification unit according to the preset classification class through an AI model to train the AI ​​model.

[0019] Preferably, the manual classification unit stores the classification results for the sample images received from the classification unit, and the AI ​​model training unit provides the classification results of the AI ​​model to the classification unit for verification and trains the AI ​​model using the verification content as feedback.

[0020] Preferably, the AI-based classification system for the unstructured data may further include an automatic classification unit that automatically classifies all images of the unstructured image data according to the preset classification class through the AI ​​model that has been trained by the AI ​​model training unit.

[0021] Preferably, the above-mentioned preset classification class can be provided as a classification class that subdivides the original classification class targeted by the AI-based classification system.

[0022] Preferably, the AI-based classification system for the above-mentioned unstructured data may further include a grouping unit that groups all images classified according to the preset classification class by the automatic classification unit according to the original classification class.

[0023] Preferably, the AI-based classification system for the unstructured data may further include an operation management unit that provides a user interface for a user to operate and manage the AI ​​model through the manual classification unit, the AI ​​model training unit, the automatic classification unit, and the grouping unit.

[0024] Preferably, the atypical image data may be provided as image data for evaluating the quality of the concrete material mix, or as image data for assessing the risk of the building by identifying the size and number of cracks that have occurred in the building.

[0025] The above objective is achieved by an AI-based classification method for unstructured data, characterized in that, according to one embodiment of the present invention, a manual classification unit of an AI-based classification system provides a classification worker to manually classify sample images extracted from unstructured image data subject to classification according to a preset classification class, and an AI model learning unit of the AI-based classification system automatically classifies the sample images classified by the manual classification unit according to the preset classification class through an AI model to train the AI ​​model.

[0026] Preferably, the AI-based classification method for the above-mentioned unstructured data may further include a step in which an automatic classification task unit of the AI-based classification system automatically classifies all images of the above-mentioned unstructured image data according to the preset classification class through the AI ​​model that has been trained by the AI ​​model training unit.

[0027] Preferably, the above-mentioned preset classification class is provided as a classification class that subdivides the original classification class targeted by the AI-based classification system, and the AI-based classification method for the above-mentioned unstructured data may further include the step of the grouping unit of the AI-based classification system grouping all images classified according to the above-mentioned preset classification class by the automatic classification unit according to the original classification class. Effects of the invention

[0029] The present invention relates to a system and method for classifying unstructured image data with ambiguous classification criteria according to a target classification class. In this system and method, a classification worker manually classifies sample images extracted from the unstructured image data subject to classification according to a pre-set classification class, and then trains an AI model using these sample images. Subsequently, the AI ​​model trained on the entire unstructured image data automatically classifies the data according to the pre-set classification class, thereby reducing the time required for classifying unstructured image data while ensuring high accuracy and reliability. Brief explanation of the drawing

[0031] FIG. 1 is a block diagram illustrating an AI-based classification system for unstructured data according to an embodiment of the present invention. Figure 2 is a schematic diagram illustrating, exemplarily, the process of the manual classification task in the AI-based classification system for unstructured data of Figure 1. Figure 3 is a schematic diagram illustrating, in an AI-based classification system for unstructured data of Figure 1, the process of the AI ​​model learning unit. Figure 4 is a schematic diagram illustrating, for example, the process of the automatic classification task unit in the AI-based classification system for unstructured data of Figure 1. Figure 5 is a schematic diagram illustrating, in terms of the operation management department, the learning process of the AI ​​model, the automatic classification process by the AI ​​model, and the grouping process in the AI-based classification system for unstructured data of Figure 1. FIG. 6 is a process flowchart illustrating an AI-based classification method for unstructured data according to an embodiment of the present invention. Specific details for implementing the invention

[0032] In order to fully understand the present invention, the operational advantages of the present invention, and the objectives achieved by the implementation of the present invention, reference must be made to the accompanying drawings illustrating preferred embodiments of the present invention and the contents described in the accompanying drawings.

[0033] The present invention will be described in detail below by explaining preferred embodiments with reference to the attached drawings. However, in describing the present invention, descriptions of already known functions or configurations will be omitted to clarify the gist of the invention.

[0035] The AI-based classification system for unstructured data according to the present invention is a system that applies an AI model to unstructured image data to classify classes according to a target classification class with high accuracy and reliability. However, the technical concept of the present invention is also applicable to an AI-based classification system for structured data.

[0036] Here, unlike structured image data, which allows for classification based on clear criteria and thus enables high training accuracy for AI models, unstructured image data lacks a consistent pattern. Consequently, classifying unstructured image data directly makes it difficult to improve training accuracy due to ambiguous classification criteria. As a result, high accuracy and reliable results cannot be expected from AI models performing classification on unstructured image data.

[0037] Preferably, the atypical image data applied in the present invention may be provided as image data for evaluating the quality of the concrete material mix (e.g., fluidity, flow, etc.) or as image data for assessing the risk of a building by identifying the size and number of cracks that have occurred in the building. However, it goes without saying that the atypical image data applied in the present invention is not limited thereto and may be provided as various other atypical image data.

[0039] FIG. 1 is a block diagram illustrating an AI-based classification system for unstructured data according to an embodiment of the present invention.

[0040] Referring to FIG. 1, an AI-based classification system (100) for unstructured data according to one embodiment of the present invention includes a manual classification unit (110), an AI model learning unit (120), an automatic classification unit (130), a grouping unit (140), and an operation management unit (150).

[0041] Figure 2 is a schematic diagram illustrating, exemplarily, the process of the manual classification task in the AI-based classification system for unstructured data of Figure 1.

[0042] Referring to FIGS. 1 and 2, the manual classification unit (110) provides the classification worker with the sample images previously extracted from the unstructured image data to be classified, so that the classification worker can manually classify them according to a pre-set classification class. Then, the manual classification unit (110) stores the classification results for the sample images received from the classification worker. At this time, the manual classification unit (110) may randomly extract sample images that are part of the total images of the unstructured image data and store them in the 'sample image directory to be classified', and store the sample images whose class has been classified by the classification worker in the 'classified sample image directory'. For example, as shown in FIG. 2, if the total number of images of the unstructured image data is 100,000, the sample images can be extracted as approximately 100.

[0043] Here, it is desirable that the classification team consist of person(s) possessing professional knowledge and experience regarding the unstructured image data that is the subject of the classification task. For example, if the unstructured image data is image data intended to evaluate the quality of concrete mix proportions, the classification team may consist of expert(s) with years of experience working at companies related to concrete mix proportions and / or professor(s) specializing in concrete mix proportions.

[0044] In addition, in this embodiment, it is preferable that the 'pre-set classification class' be provided as a classification class that subdivides the original classification class targeted by the AI-based classification system (100) for unstructured data in order to further improve the accuracy and reliability of the classification result. That is, the pre-set classification class is a classification class set by subdividing the original classification class targeted according to certain criteria, and can be composed of a larger number of classes than the number of classes of the original classification class.

[0045] For example, as illustrated in FIG. 2, if the target original classification class consists of three classes, 'Class_OK', 'Class_NA', and 'Class_NOK', the preset classification class may subdivide the original classification class consisting of these three classes into ten classes, 'Class_1', 'Class_2', 'Class_3', 'Class_4', 'Class_5', 'Class_6', 'Class_7', 'Class_8', 'Class_9', and 'Class_10'. For reference, in the original classification class, 'Class_OK', 'Class_NA', and 'Class_NOK' may mean 'Good', 'Classification Pending', and 'Bad', respectively.

[0046] However, unlike the present embodiment, the 'pre-set classification class' in the present invention may be provided as the original classification class targeted by the AI-based classification system for unstructured data. That is, in the present invention, the 'pre-set classification class' may be provided as a classification class that subdivides the original classification class targeted by the AI-based classification system for unstructured data, or it may be provided as the original classification class itself.

[0047] Figure 3 is a schematic diagram illustrating, in an AI-based classification system for unstructured data of Figure 1, the process of the AI ​​model learning unit.

[0048] Referring to FIGS. 1 and 3, the AI ​​model learning unit (120) automatically classifies sample images classified according to a classification class previously set by the manual classification unit (110) according to a classification class set by the AI ​​model (10). Then, the AI ​​model learning unit (120) provides the classification results of the AI ​​model (10) to the classification worker for verification, and trains the AI ​​model (10) using the verification results as feedback. This training process of the AI ​​model (10) by the AI ​​model learning unit (120) can be repeated until a target learning accuracy (e.g., 98% accuracy) is achieved. At this time, the AI ​​model learning unit (120) can store sample images whose classes are manually classified by the manual classification unit (110) in the 'sample image directory to be classified (120A, see FIG. 5)' and sample images whose classes are automatically classified by the AI ​​model (10) in the 'classified sample image directory (120B, see FIG. 5)'.

[0049] For example, as illustrated in FIG. 3, the AI ​​model learning unit (120) can input 100 sample images classified into 10 classes by the manual classification unit (110) into the AI ​​model (10) for the training of the AI ​​model (10), output the results of the AI ​​model (10) classifying the 100 sample images into 10 classes, and provide this to the classifier. Then, the AI ​​model learning unit (120) can retrain the AI ​​model (10) using the content verified by the classification worker regarding the classification result of the AI ​​model (10) as feedback, and this training process of the AI ​​model (10) can be performed repeatedly until the target accuracy is achieved. Here, the classification worker can verify the classification result of the AI ​​model (10) by directly looking at it, and if there is an incorrect classification result, correct it and classify it, and then input the verification content into the AI ​​model learning unit (120).

[0050] Figure 4 is a schematic diagram illustrating, for example, the process of the automatic classification task unit in the AI-based classification system for unstructured data of Figure 1.

[0051] Referring to FIGS. 1 and 4, the automatic classification unit (130) automatically classifies all images of unstructured image data according to a pre-set classification class through an AI model (10) that has been trained by the AI ​​model training unit (120), that is, has achieved the target training accuracy. In other words, the automatic classification unit (130) inputs all images of unstructured image data into the AI ​​model (10), rather than sample images extracted from a portion of the unstructured image data, and outputs the result of the AI ​​model (10) automatically classifying all images according to a pre-set classification class. At this time, the automatic classification unit (130) may store all images of unstructured image data in a 'directory of all images to be classified (130A, see FIG. 5)' and store all images whose classes have been automatically classified by the trained AI model (10) in a 'directory of all classified images (130B, see FIG. 5)'.

[0052] For example, as illustrated in FIG. 4, the automatic classification unit (130) inputs 1 million unstructured images of unstructured image data into a trained AI model (10), and outputs the result of the AI ​​model (10) classifying the 1 million unstructured images into 10 classes.

[0053] Referring to FIGS. 1 and 4, the grouping unit (140) groups the entire images classified according to a classification class previously set by the automatic classification unit (130) according to the original classification class targeted by the AI-based classification system (100) for unstructured data. That is, the grouping unit (140) performs the task of converting the original classification class, which has a larger number of classes due to subdivision, into a pre-set classification class that has a smaller number of classes.

[0054] For example, as illustrated in FIG. 4, the grouping unit (140) can group the 10 classes, consisting of 'Class_1', 'Class_2', 'Class_3', 'Class_4', 'Class_5', 'Class_6', 'Class_7', 'Class_8', 'Class_9' and 'Class_10', which are classified according to a classification class preset by the automatic classification unit (130), into 3 classes consisting of 'Class_OK', 'Class_NA' and 'Class_NOK' according to the original classification class. At this time, regarding the entire set of images, as shown in FIG. 4, images corresponding to 'Class_1', 'Class_2', 'Class_3' and 'Class_4' respectively are grouped and classified as 'Class_NOK', images corresponding to 'Class_5' and 'Class_6' respectively are grouped and classified as 'Class_NA', and images corresponding to 'Class_7', 'Class_8', 'Class_9' and 'Class_10' respectively are grouped and classified as 'Class_OK'.

[0055] Figure 5 is a schematic diagram illustrating, in terms of the operation management department, the learning process of the AI ​​model, the automatic classification process by the AI ​​model, and the grouping process in the AI-based classification system for unstructured data of Figure 1.

[0056] Referring to FIGS. 1 and 5, the operation management unit (150) provides a user interface for the user to operate and manage the AI ​​model (10) through the previously described manual classification unit (110), AI model learning unit (120), automatic classification unit (130), and grouping unit (140).

[0057] For example, the operation management unit (150) may request training and retraining of the AI ​​model (10) as shown in FIG. 5 and display the training accuracy and real-time progress status of the AI ​​model (10) accordingly, or request automatic classification of all images from the AI ​​model (10) and display the real-time progress status and estimated completion time accordingly, or request class grouping according to the original classification class for the automatic classification results of the AI ​​model (10) and display the results and estimated completion time, and the user may input commands or instructions for necessary tasks through the user interface provided by the operation management unit (150) and receive the output results through the user interface.

[0058] At this time, in the learning process of the AI ​​model (10), as shown in FIG. 5, sample images stored in the 'sample image directory to be classified' (120A) are automatically fed into the AI ​​model (10), and the classification result of the AI ​​model (10) for the sample images can be stored in the 'classified sample image directory' (120B). Also, in the automatic classification process of the AI ​​model (10), as shown in FIG. 5, all images stored in the 'all image directory to be classified' (130A) are automatically fed into the AI ​​model (10), and the classification result of the AI ​​model (10) for all images can be stored in the 'classified all image directory' (130B).

[0060] FIG. 6 is a process flowchart illustrating an AI-based classification method for unstructured data according to an embodiment of the present invention.

[0061] The AI-based classification method for unstructured data described herein is merely one embodiment of the present invention, and various additional steps may be added as needed. Furthermore, since the steps below may be performed in a different order, the present invention is not limited to each step and its order described below.

[0062] Referring to FIGS. 1 and FIGS. 6, an AI-based classification method for unstructured data according to one embodiment of the present invention is a method executed by an AI-based classification system (100) and includes a manual classification step (S110), an AI model learning step (S120), an automatic classification step (S130), and a grouping step (S140).

[0063] First, in the manual classification step (S110), the manual classification task unit (110) of the AI-based classification system (100) provides the classification worker with the sample images extracted from the unstructured image data to be classified according to a pre-set classification class. Here, the 'pre-set classification class' is provided as a classification class that subdivides the 'original classification class' targeted by the AI-based classification system (100).

[0064] Next, in the AI ​​model learning step (S120), the AI ​​model learning unit (120) of the AI-based classification system (100) automatically classifies sample images classified by the manual classification unit (110) according to a pre-set classification class through the AI ​​model (10) to train the AI ​​model (10). This AI model learning step (S120) is repeated until a target learning accuracy (e.g., 98% accuracy) is achieved, as illustrated in FIG. 6.

[0065] Next, in the automatic classification step (S130), the automatic classification task unit (130) of the AI-based classification system (100) automatically classifies all images of the unstructured image data according to a pre-set classification class through the AI ​​model (10) that has been trained by the AI ​​model training unit (120).

[0066] Next, in the grouping step (S140), the grouping work unit (140) of the AI-based classification system (100) groups all images classified according to a classification class pre-set by the automatic classification work unit (130) according to the original classification class.

[0068] It is obvious to those skilled in the art that the present invention is not limited to the embodiments described above and can be modified and varied in various ways without departing from the spirit and scope of the invention. Accordingly, such modifications or variations should be deemed to fall within the scope of the claims of the present invention. Explanation of the symbols

[0070] 100: AI-based classification system for unstructured data 10: AI Model 110: Manual sorting unit 120: AI model training unit 130: Automatic sorting unit 140: Grouping Task 150: Operations Management Department

Claims

Claim 1 A manual classification unit that provides a classification worker with the ability to manually classify sample images extracted from unstructured image data subject to classification work according to a preset classification class; and an AI model training unit that automatically classifies the sample images classified by the manual classification unit according to the preset classification class through an AI model and trains the AI ​​model. An AI-based classification system for unstructured data comprising: an automatic classification unit that automatically classifies all images of the unstructured image data according to a preset classification class through the AI ​​model trained by the AI ​​model training unit; wherein, in terms of improving the accuracy and reliability of the classification results of the AI-based classification system, the preset classification class is provided as a classification class composed of a greater number of classes than the original classification class by subdividing the original classification class targeted by the AI-based classification system, and the original classification class includes three classes meaning 'Good', 'Classification Pending', and 'Bad'; and further comprising a grouping unit that groups all images classified according to the preset classification class by the automatic classification unit according to the original classification class; wherein the grouping unit performs a task of converting the preset classification class, which has a greater number of classes than the original classification class due to the subdivision of the original classification class, into the original classification class having a smaller number of classes. Claim 2 An AI-based classification system for unstructured data according to claim 1, wherein the manual classification unit stores the classification results for the sample images received from the classification worker, and the AI ​​model learning unit provides the classification results of the AI ​​model to the classification worker for verification and trains the AI ​​model using the verification content as feedback. Claim 3 delete Claim 4 delete Claim 5 delete Claim 6 An AI-based classification system for unstructured data according to claim 1, further comprising an operation management unit that provides a user interface for a user to operate and manage an AI model through the manual classification unit, the AI ​​model learning unit, the automatic classification unit, and the grouping unit. Claim 7 An AI-based classification system for unstructured data according to claim 1, characterized in that the unstructured image data is provided as image data for evaluating the quality of a concrete material mix, or as image data for assessing the risk of a building by identifying the size and number of cracks that have occurred in the building. Claim 8 delete

Images