With the recent surge in YouTube usage, there has been a proliferation of user-generated videos where individuals evaluate cosmetics. Consequently, many companies are increasingly utilizing evaluation videos for their product marketing and market research. However, a notable drawback is the manual classification of these product review videos incurring significant costs and time. Therefore, this paper proposes a deep learning-based cosmetics search algorithm to automate this task. The algorithm consists of two networks: One for detecting candidates in images using shape features such as circles, rectangles, etc and Another for filtering and categorizing these candidates. The reason for choosing a Two-Stage architecture over One-Stage is that, in videos containing background scenes, it is more robust to first detect cosmetic candidates before classifying them as specific objects. Although Two-Stage structures are generally known to outperform One-Stage structures in terms of model architecture, this study opts for Two-Stage to address issues related to the acquisition of training and validation data that arise when using One-Stage. Acquiring data for the algorithm that detects cosmetic candidates based on shape and the algorithm that classifies candidates into specific objects is cost-effective, ensuring the overall robustness of the algorithm.

The color image of the brand comes first and is an important visual element that leads consumers to the consumption of the product. To express more effectively what the brand wants to convey through design, the printing market is striving to print accurate colors that match the intention. In ‘offset printing’ mainly used in printing, colors are often printed in CMYK (Cyan, Magenta, Yellow, Key) colors. However, it is possible to print more accurate colors by making ink of the desired color instead of dotting CMYK colors. The resulting ink is called ‘spot color’ ink. Spot color ink is manufactured by repeating the process of mixing the existing inks. In this repetition of trial and error, the manufacturing cost of ink increases, resulting in economic loss, and environmental pollution is caused by wasted inks. In this study, a deep learning algorithm to predict printed spot colors was designed to solve this problem. The algorithm uses a single DNN (Deep Neural Network) model to predict printed spot colors based on the information of the paper and the proportions of inks to mix. More than 8,000 spot color ink data were used for learning, and all color was quantified by dividing the visible light wavelength range into 31 sections and the reflectance for each section. The proposed algorithm predicted more than 80% of spot color inks as very similar colors. The average value of the calculated difference between the actual color and the predicted color through ‘Delta E’ provided by CIE is 5.29. It is known that when Delta E is less than 10, it is difficult to distinguish the difference in printed color with the naked eye. The algorithm of this study has a more accurate prediction ability than previous studies, and it can be added flexibly even when new inks are added. This can be usefully used in real industrial sites, and it will reduce the attempts of the operator by checking the color of ink in a virtual environment. This will reduce the manufacturing cost of spot color inks and lead to improved working conditions for workers. In addition, it is expected to contribute to solving the environmental pollution problem by reducing unnecessarily wasted ink.

본 논문은 딥러닝 알고리즘을 이용하여 딸기 영상 데이터의 병충해 존재 여부를 자동으로 검출할 수 있는 서비스 모델을 제안한다. 또한 병징에 특화된 분할 이미지 데이터 세트를 제 안하여 딥러닝 모델의 병충해 검출 성능을 향상한다. 딥러닝모델은 CNN 기반 YOLO를 선정하여 기존의 R-CNN 기반 모델의 느린 학습속도와 추론속도를 개선하였다. 병충해 검 출 모델을 학습하기 위해 일반적인 데이터 세트와 제안하는 분할 이미지 데이터 세트를 구축하였다. 딥러닝 모델이 일반 적인 학습 데이터 세트를 학습했을 때 병충해 검출률은 81.35%이며 병충해 검출 신뢰도는 73.35%이다. 반면 딥러닝 모델이 분할 이미지 학습 데이터 세트를 학습했을 때 병충해 검출률은 91.93%이며 병충해 검출 신뢰도는 83.41%이다. 따 라서 분할 이미지 데이터를 학습한 딥러닝 모델의 성능이 우 수하다는 것을 증명할 수 있었다.

It is important to ensure worker’s safety from radiation hazard in decommissioning site. Real-time tracking of worker’s location is one of the factors necessary to detect radiation hazard in advance. In this study, the integrated algorithm for worker tracking has been developed to ensure the safety of workers. There are three essential techniques needed to track worker’s location, which are object detection, object tracking, and estimating location (stereo vision). Above all, object detection performance is most important factor in this study because the performance of tracking and estimating location is depended on worker detection level. YOLO (You Only Look Once version 5) model capable of real-time object detection was applied for worker detection. Among the various YOLO models, a model specialized for person detection was considered to maximize performance. This model showed good performance for distinguishing and detecting workers in various occlusion situations that are difficult to detect correctly. Deep SORT (Simple Online and Realtime Tracking) algorithm which uses deep learning technique has been considered for object tracking. Deep SORT is an algorithm that supplements the existing SORT method by utilizing the appearance information based on deep learning. It showed good tracking performance in the various occlusion situations. The last step is to estimate worker’s location (x-y-z coordinates). The stereo vision technique has been considered to estimate location. It predicts xyz location using two images obtained from stereo camera like human eyes. Two images are obtained from stereo camera and these images are rectified based on camera calibration information in the integrated algorithm. And then workers are detected from the two rectified images and the Deep SORT tracks workers based on worker’s position and appearance between previous frames and current frames. Two points of workers having same ID in two rectified images give xzy information by calculating depth estimation of stereo vision. The integrated algorithm developed in this study showed sufficient possibility to track workers in real time. It also showed fast speed to enable real-time application, showing about 0.08 sec per two frames to detect workers on a laptop with high-performance GPU (RTX 3080 laptop version). Therefore, it is expected that this algorithm can be sufficiently used to track workers in real decommissioning site by performing additional parameter optimization.

본 연구는 화재진압 및 피난활동을 지원하는 딥러닝 기반의 알고리즘 개발에 관한 기초 연구로 선박 화재 시 연기감지기가 작동하기 전에 검출된 연기 데이터를 분석 및 활용하여 원격지까지 연기가 확산 되기 전에 연기 확산거리를 예측하는 것이 목적이다. 다음과 같은 절차에 따라 제안 알고리즘을 검토하였다. 첫 번째 단계로, 딥러닝 기반 객체 검출 알고리즘인 YOLO(You Only Look Once)모델에 화재시뮬레이션을 통하여 얻은 연기 영상을 적용하여 학습을 진행하였다. 학습된 YOLO모델의 mAP(mean Average Precision)은 98.71%로 측정되었으며, 9 FPS(Frames Per Second)의 처리 속도로 연기를 검출하였다. 두 번째 단계로 YOLO로부터 연기 형상이 추출된 경계 상자의 좌표값을 통해 연기 확산거리를 추정하였으며 이를 시계열 예측 알고리즘인 LSTM(Long Short-Term Memory)에 적용하여 학습을 진행하였다. 그 결과, 화재시뮬레이션으로부터 얻은 Fast 화재의 연기영상에서 경계 상자의 좌표값으로부터 추정한 화재발생~30초까지의 연기 확산거리 데이터를 LSTM 학습모델에 입력하여 31초~90초까지의 연기 확산거리 데이터를 예측하였다. 그리고 추정한 연기 확산거리와 예측한 연기 확산거리의 평균제곱근 오차는 2.74로 나타났다.

본 연구에서는 영상기반 딥러닝 및 이미지 프로세싱 기법을 이용한 볼트풀림 손상검출 기법을 제안하였다. 이를 위해 먼저, 딥러닝 및 이미지 프로세싱 기반 볼트풀림 검출 기법을 설계하였다. 영상기반 볼트풀림 검출 기법은 볼트 이미지 검출 과정 및 볼트풀림 각도 추정 과정으로 구성된다. 볼트 이미지의 검출을 위하여 RCNN기반 딥러닝 알고리즘을 이용하였다. 영상의 원근왜곡 교정을 위해 호모그래피 개념을 이용하였으며 볼트풀림 각도를 추정을 위하여 Hough 변환을 이용하였다. 다음으로 제안된 기법의 성능을 검증을 위하여 거더의 볼트 연결부 모형을 대상으로 볼트풀림 손상검출 실험을 수행하였다. 다양한 원근 왜곡 조건에 대하여 RCNN 기반 볼트 검출기와 Hough 변환 기반 볼트풀림 각도 추정기의 성능을 검토하였다.

The incidence of stomach cancer has been found to be gradually decreasing; however, it remains one of the most frequently occurring malignant cancers in Korea. According to statistics of 2017, stomach cancer is the top cancer in men and the fourth most important cancer in women, necessitating methods for its early detection and treatment. Considerable research in the field of bioinformatics has been conducted in cancer studies, and bioinformatics approaches might help develop methods and models for its early prediction. We aimed to develop a classification method based on deep learning and demonstrate its application to gene expression data obtained from patients with stomach cancer. Data of 60,483 genes from 334 patients with stomach cancer in The Cancer Genome Atlas were evaluated by principal component analysis, heatmaps, and the convolutional neural network (CNN) algorithm. We combined the RNA-seq gene expression data with clinical data, searched candidate genes, and analyzed them using the CNN deep learning algorithm. We performed learning using the sample type and vital status of patients with stomach cancer and verified the results. We obtained an accuracy of 95.96% for sample type and 50.51% for vital status. Despite overfitting owing to the limited number of patients, relatively accurate results for sample type were obtained. This approach can be used to predict the prognosis of stomach cancer, which has many types and underlying causes.

딥러닝 모델은 주어진 학습용 데이터에서 탐지하고자 하는 물체의 특징을 추출하기 때문에, 딥러닝 모델 학습을 위한 학습용 데이터 구축은 매우 중요하다. 본 연구에서는 균열을 탐지하는 딥러닝 모델의 성능을 향상시키기 위해, 실제 콘크리트 구조물이나 아스팔트 도로 표면에서 자주 발견될 수 있는 나뭇가지, 거미줄, 전선 등을 학습 데이터에 자동으로 포함시키고, negative 영역으로 분류하는 알고리즘을 개발하였다. 제안된 알고리즘을 사용하여 학습된 딥러닝 모델을 실제 도로 표면에 발생한 균열 탐지에 적용하여 실제 균열 탐지에 사용될 수 있음을 보였다.

Carbonation of reinforced concrete is a major factor in the deterioration of reinforced concrete, and prediction of the resistance to carbonation is important in determining the durability life of reinforced concrete structures. In this study, basic research on the prediction of carbonation penetration depth of concrete using Deep Learning algorithm among artificial neural network theory was carried out. The data used in the experiment were analyzed by deep running algorithm by setting W/B, cement and blast furnace slag, fly ash content, relative humidity of the carbonated laboratory, temperature, CO2 concentration, Deep learning algorithms were used to study 60,000 times, and the analysis of the number of hidden layers was compared.

최근 기후변화 및 유역개발로 인하여 메콩강 유역의 수문환경이 급격히 변화하고 있으며, 메콩강을 공유하는 국가의 수재해 예방 및 지속가능한 수자원개발을 위해서는 메콩강 주요지점에서의 유량 정보의 분석 및 예측이 요구된다. 본 연구에서는 물리적 기반의 수문모형인 SWAT과 데이터기반 딥러닝 알고리즘인 LSTM을 이용하여 메콩강 하류 Kratie 지점의 유출모의를 수행하고, 유출모의 정확도 및 두 가지 방법론의 장 ․ 단점을 비교 ․ 분석한다. SWAT 모형의 구축을 위해 범용 입력자료(지형: HydroSHED, 토지이용: GLCF-MODIS, 토양: FAO-Soil map, 강우: APHRODITE 등)을 이용하였으며 warming-up 및 매개변수 보정 후 2003~2007년 일유량 모의를 수행하였다. LSTM을 이용한 유출모의의 경우, 딥러닝 오픈소스 라이브러리인 TensorFlow를 활용하여 Kratie 지점기준 메콩강 상류 10개 수위관측소의 두 기간(2000~2002, 2008~2014) 일수위 정보만을 이용하여 심층신경망을 학습하고, SWAT 모형과 마찬가지로 2003~2007년을 대상으로 Kratie 지점에 대한 일수위 모의 후 수위-유량관계곡선식을 이용하여 유출량으로 환산하였다. 두 모형의 모의성능 비교 ․ 검토를 위하여 모의기간에 대해 NSE (Nash-Sutcliffe Efficiency)을 산정한 결과, SWAT은 0.9, LSTM은 보다 높은 0.99의 정확도를 나타내는 것으로 분석되었다. 메콩강과 같은 대유역의 특정 지점에 대한 수문시계열 자료의 모의를 위해서는 다양한 입력자료를 요구하는 물리적 수문모형 대신 선행시계열자료의 변동성을 기억 ․ 학습하여 이를 예측에 반영하는 LSTM 기법 등 데이터기반의 심층신경망 모형의 적용이 가능할 것으로 판단된다.

The conventional method for estimating compressive strength of concrete has been suggested by considering only 1 to 3 influential factors. In this study, seven influential mixture factors (Water-Cement Ratio, Water, Cement, Fly ash, Blast furnace slag, Curing temperature, and humidity) of papers opened for 10 years were collected at three conferences in order to know tendency of data. The purpose of this paper is to estimate compressive strength more accurately by applying it to algorithm of the Deep learning.

As the importance of maintenance of reinforced concrete structures spreads, interest in the durability of structures is increasing. Among them, carbonation of concrete is one of the main deterioration factors of reinforced concrete structures. For quantitative evaluation of carbonation, many researchers are predicting carbonation considering water-cement ratio and environmental requirements. In this study, we studied the parameters based on the concrete made of ordinary Portland cement in the existing experimental data. The depth of carbonation deduced from the learning is applied to the carbonation by applying the deep learning.