• Composites Research
• /
• v.36 no.6
• /
• pp.395-401
• /
• 2023

In this study, finite element modeling of unidirectional composite materials of the computed tomography (CT) was conducted using a supervised learning-based segmentation technique. Firstly, Micro-CT scan was performed to obtain the raw volume of unidirectional composite materials, providing microstructure information. From the CT volume images, actual microstructure of the cross-section of unidirectional composite materials was extracted by the labeling process. Then, a U-net deep learning model was trained with a small number of raw images as inputs and their labeled images as outputs to generate a segmentation model. Subsequently, most of remaining images were input to the trained U-net deep learning model to segment all raw volume for identifying complex microstructure, which was used for the generation of finite element model. Finally, the fiber volume fraction of the finite element model was compared with that of experimentally measured volume to validate the appropriateness of the proposed method.

• Journal of Platform Technology
• /
• v.11 no.3
• /
• pp.56-67
• /
• 2023

Recently, de-identification of personal information, which has been a long-cherished desire of the data-based industry, was revised and specified in August 2020. It became the foundation for activating data called crude oil[2] in the fourth industrial era in the industrial field. However, some people are concerned about the infringement of the basic rights of the data subject[3]. Accordingly, a development study was conducted on the Batch De-Identification Tool, a personal information de-identification automation tool. In this study, first, we developed an image labeling tool to label human faces (eyes, nose, mouth) and car license plates of various resolutions to build data for training. Second, an object recognition model was trained to run the object recognition module to perform de-identification of personal information. The automated personal information de-identification tool developed as a result of this research shows the possibility of proactively eliminating privacy violations through online services. These results suggest possibilities for data-based industries to maximize the value of data while balancing privacy and utilization.

• Journal of the Korea Society of Computer and Information
• /
• v.28 no.5
• /
• pp.17-28
• /
• 2023

Potholes that occur on paved roads can have fatal consequences for vehicles traveling at high speeds and may even lead to fatalities. While manual detection of potholes using human labor is commonly used to prevent pothole-related accidents, it is economically and temporally inefficient due to the exposure of workers on the road and the difficulty in predicting potholes in certain categories. Therefore, completely preventing potholes is nearly impossible, and even preventing their formation is limited due to the influence of ground conditions closely related to road environments. Additionally, labeling work guided by experts is required for dataset construction. Thus, in this paper, we utilized the Mean Teacher technique, one of the semi-supervised learning-based knowledge distillation methods, to achieve robust performance in pothole image classification even with limited labeled data. We demonstrated this using performance metrics and GradCAM, showing that when using semi-supervised learning, 15 pre-trained CNN models achieved an average accuracy of 90.41%, with a minimum of 2% and a maximum of 9% performance difference compared to supervised learning.

• Journal of Broadcast Engineering
• /
• v.24 no.1
• /
• pp.67-76
• /
• 2019

In this paper, we propose an efficient object classification method based on semantic segmentation for multi-labeled image data. In addition to various pixel unit information and processing techniques such as color information, contour, contrast, and saturation included in image data, a detailed region in which each object is located is extracted as a meaningful unit and the experiment is conducted to reflect the result in the inference. We use a neural network that has been proven to perform well in image classification to understand which object is located where image data containing various class objects are located. Based on these researches, we aim to provide artificial intelligence services that can classify real-time detailed areas of complex images containing various objects in the future.

• Proceedings of the Korean Institute of Information and Commucation Sciences Conference
• /
• 2022.05a
• /
• pp.154-157
• /
• 2022

In the existing Tensorflow CNN environment, the object detection method is a method of performing object labeling and detection by Tensorflow itself. However, with the advent of YOLO, the efficiency of image object detection has increased. As a result, more deep layers can be built than existing neural networks, and the image object recognition rate can be increased. Therefore, in this paper, the detection ability and speed were compared and analyzed by designing an object detection system based on Darknet and YOLO and performing multi-layer construction and learning based on the existing convolutional neural network. For this reason, in this paper, a neural network methodology that efficiently uses Darknet's learning is presented.

• Proceedings of the Korean Society of Computer Information Conference
• /
• 2022.07a
• /
• pp.45-48
• /
• 2022

건축물 안전 점검은 대부분 전문가의 현장 방문을 통한 육안검사다. 그중 균열 검사는 건물 위험도를 나타내는 중요한 지표로써 발생 위치, 진행성, 크기를 조사하는데, 최근 균열 조사 방식에 대해 객관성과 체계성을 보완할 딥러닝 개발이 활발하다. 그러나 균열 이미지는 외부 현장에 모양, 규모도 많은 종류라 도메인이 다양해야 하는데 대부분 제한된 환경과 실제적인 균열 검사와는 무관한 데이터로 구성되어 실효적이지 않다. 본 연구에서는 균열 조사에 적합하고 Wild 환경에 적용 가능한 POC 데이터셋을 소개한다. 기존 균열 공인 데이터셋 4종의 특징과 한계점을 분석을 토대로 고해상도 이미지로써 균열의 세부 특징을 담았고 균열 유사 환경과 조건들을 추가 촬영해 균열 검출에 강인하게 학습되도록 지향하였다. 정제 및 라벨링 작업을 거친 POC 데이터 셋은 균열 검출모델인 YOLO-v5으로 성능을 실험하였고, mAP(mean Average Precision) 75.5%로 높은 검출률을 보였다. POC 데이터셋으로 더욱 도메인에 적응적(Domain-adapted)인 인공지능 모델을 개발하여 건물, 댐, 교량 등 각종 대형 건축물에 대한 안전하고 효과적인 안전 관리 도구로써 활용할 것을 기대한다.

• Journal of Drive and Control
• /
• v.20 no.4
• /
• pp.92-99
• /
• 2023

Autonomous vehicles should recognize and respond to the specified speed to drive in compliance with regulations. To recognize the specified speed, the most representative method is to read the numbers of the signs by recognizing the speed signs in the front camera image. This study proposes a method that utilizes YOLO-Labeling-Labeling-EfficientNet. The sign box is first recognized with YOLO, and the numeric digit is extracted according to the pixel value from the recognized box through two labeling stages. After that, the number of each digit is recognized using EfficientNet (CNN) learned with the virtual environment dataset produced directly. In addition, we estimated the depth of information from the height value of the recognized sign through regression analysis. We verified the proposed algorithm using the virtual racing environment and GTSRB, and proved its real-time performance and efficient recognition performance.

• Proceedings of the Korean Society of Broadcast Engineers Conference
• /
• 2022.11a
• /
• pp.183-186
• /
• 2022

최근 인공지능 기반 객체 탐지 기술이 발전함에 따라 영상 감시, 얼굴 인식, 로봇 제어, IoT, 자율주행, 제조업, 보안 등 다양한 분야에 활용되고 있다. 이에 본 논문은 발전된 객체 탐지 알고리즘을 이용하여 비전문가에겐 생소한 컴퓨터나 전기 장치 등의 '단자(terminal)' 모양을 구별하는 방법을 제안한다. 이를 위해 객체 탐지 프로그램인 You Only Look Once (YOLO) 알고리즘을 이용하여 입력한 단자들의 모양을 검출하는 알고리즘을 구성하였다. 일상에서 쉽게 볼 수 있는 단자들의 이미지(VGA, DVI, HDMI, DP, USB-A, USB-C)를 라벨링하여 데이터셋을 구축하였고, YOLOv4와 YOLOv5 두 버전의 알고리즘을 사용하여 성능을 검증하였다. 실험 결과 mean Average Precision(mAP) 기준 최대 92.9%의 정확도를 얻을 수 있었다. 전기 장치에 따라 단자의 모양이 다양하고, 그 종류 또한 많기 때문에 본 연구가 방송 기술 등의 여러 분야에 응용될 것으로 기대된다.

• Journal of IKEEE
• /
• v.25 no.4
• /
• pp.613-618
• /
• 2021

Facial expression recognition has long been established as a subject of continuous research in various fields. In this paper, the relationship between each landmark is analyzed using the features obtained by calculating the distance between the facial landmarks in the image, and five facial expressions are classified. We increased data and label reliability based on our labeling work with multiple observers. In addition, faces were recognized from the original data and landmark coordinates were extracted and used as features. A genetic algorithm was used to select features that are relatively more helpful for classification. We performed facial recognition classification and analysis with the method proposed in this paper, which shows the validity and effectiveness of the proposed method.

• Proceedings of the Korean Institute of Information and Commucation Sciences Conference
• /
• 2021.05a
• /
• pp.165-168
• /
• 2021

Recently, the spread of deep learning environments has increased the importance of dataset generation. In this paper, we aim to design and implement a method for capturing rotating images of objects and performing object tracking on them for efficient dataset generation. The method implemented in this paper is to obtain image data by rotating objects to capture multiple angles of objects, detect and track objects through background removal and difference image processing techniques, showing them on screen to monitor object tracking results in the current frame. It was then implemented to return object location data within the image for use as a dataset.