• 문화기술의 융합
• /
• 제6권4호
• /
• pp.789-795
• /
• 2020

전 세계적으로 유행하는 코로나19로 인해 많은 사망자가 보고되고 있다. 코로나19의 추가 확산을 막기 위해서는 의심 환자에 대해 신속하고 정확한 영상판독을 한 후, 적절한 조치를 취해야 한다. 이를 위해 본 논문은 환자의 감염 여부를 의료진에게 제공해 영상판독을 보조할 수 있는 딥 러닝 기반 코로나19 흉부 X선 판독 기법을 소개한다. 우선 판독모델을 학습하기 위해서는 충분한 데이터셋이 확보되어야 하는데, 현재 제공하는 코로나19 오픈 데이터셋은 학습의 정확도를 보장하기에 그 영상 데이터 수가 충분하지 않다. 따라서 누적 적대적 생성 신경망(StackGAN++)을 사용해 인공지능 학습 성능을 저하하는 영상 데이터 수적 불균형 문제를 해결하였다. 다음으로 판독모델 개발을 위해 증강된 데이터셋을 사용하여 DenseNet 기반 분류모델 학습을 진행하였다. 해당 분류모델은 정상 흉부 X선과 코로나 19 흉부 X선 영상을 이진 분류하는 모델로, 실제 영상 데이터 일부를 테스트데이터로 사용하여 모델의 성능을 평가하였다. 마지막으로 설명 가능한 인공지능(eXplainable AI, XAI) 중 하나인 Grad-CAM을 사용해 입력 영상의 질환유무를 판단하는 근거를 제시하여 모델의 신뢰성을 확보하였다.

• 지능정보연구
• /
• 제29권1호
• /
• pp.41-63
• /
• 2023

전자상거래 시장의 꾸준한 성장으로 인해 추천 시스템의 필요성은 점차 강조되고 있으며, 최근에는 추천 성능의 향상을 목적으로 리뷰 텍스트를 사용하는 연구가 활발히 진행되고 있다. 특히 많은 연구들은 리뷰 텍스트의 감성 점수를 활용하여 제안되고 있는데, 감성 점수만을 사용하는 방법론은 리뷰 텍스트에 존재하는 구체적인 선호도 정보의 활용 측면에 한계를 가지며 이는 결과적으로 성능 향상에 제약으로 작용하게 된다. 이를 개선하기 위해 본 연구는 딥러닝 기반 추천 모델에 온라인 리뷰 내 다양한 언어학적 요소들을 활용하여 고객의 선호도를 정교하게 학습할 수 있는 새로운 추천 방법론을 제안하였다. 이를 위해 먼저 고객과 상품 간 복잡한 상호작용을 고려할 수 있도록 딥러닝 모델을 통해 상호작용 관계를 비선형으로 학습하였다. 그리고 리뷰 텍스트를 효과적으로 활용할 수 있도록 언어학적 요소 중 고객의 구매 의사결정에 중요한 영향을 미치는 인지적 요인, 정서적 요인 그리고 언어 스타일 매칭을 사용하였다. 실험은 Amazon.com에서 수집한 온라인 리뷰 데이터를 사용하여 진행하였고, 실험 결과 제안 모델의 우수함을 검증할 수 있었다. 본 연구는 추천 시스템에서 리뷰 텍스트 내 고객 선호도에 대한 정보를 효과적으로 활용하는 방법론을 제안하여 연구의 이론적 및 방법론 측면에 기여하였다.

• 한국안전학회지
• /
• 제36권3호
• /
• pp.31-39
• /
• 2021

In Korea, more than half of work-related fatalities have occurred on construction sites. To reduce such occupational accidents, safety inspection by government agencies is essential in construction sites that present a high risk of serious accidents. To address this issue, this study developed risk prediction models of serious accidents in construction sites using five machine learning methods: support vector machine, random forest, XGBoost, LightGBM, and AutoML. To this end, 15 proactive information (e.g., number of stories and period of construction) that are usually available prior to construction were considered and two over-sampling techniques (SMOTE and ADASYN) were used to address the problem of class-imbalanced data. The results showed that all machine learning methods achieved 0.876~0.941 in the F1-score with the adoption of over-sampling techniques. LightGBM with ADASYN yielded the best prediction performance in both the F1-score (0.941) and the area under the ROC curve (0.941). The prediction models revealed four major features: number of stories, period of construction, excavation depth, and height. The prediction models developed in this study can be useful both for government agencies in prioritizing construction sites for safety inspection and for construction companies in establishing pre-construction preventive measures.

• 산업경영시스템학회지
• /
• 제44권4호
• /
• pp.12-22
• /
• 2021

This article suggests the machine learning model, i.e., classifier, for predicting the production quality of free-machining 303-series stainless steel(STS303) small rolling wire rods according to the operating condition of the manufacturing process. For the development of the classifier, manufacturing data for 37 operating variables were collected from the manufacturing execution system(MES) of Company S, and the 12 types of derived variables were generated based on literature review and interviews with field experts. This research was performed with data preprocessing, exploratory data analysis, feature selection, machine learning modeling, and the evaluation of alternative models. In the preprocessing stage, missing values and outliers are removed, and oversampling using SMOTE(Synthetic oversampling technique) to resolve data imbalance. Features are selected by variable importance of LASSO(Least absolute shrinkage and selection operator) regression, extreme gradient boosting(XGBoost), and random forest models. Finally, logistic regression, support vector machine(SVM), random forest, and XGBoost are developed as a classifier to predict the adequate or defective products with new operating conditions. The optimal hyper-parameters for each model are investigated by the grid search and random search methods based on k-fold cross-validation. As a result of the experiment, XGBoost showed relatively high predictive performance compared to other models with an accuracy of 0.9929, specificity of 0.9372, F₁-score of 0.9963, and logarithmic loss of 0.0209. The classifier developed in this study is expected to improve productivity by enabling effective management of the manufacturing process for the STS303 small rolling wire rods.

• 디지털융복합연구
• /
• 제20권5호
• /
• pp.39-46
• /
• 2022

교통사고와 사회·경제적 손실 간의 연계성이 확인됨에 따라 사고 데이터에 기반을 둔 안전 정책 마련 및 중상·사망 등 그 심각도가 높은 교통사고의 절감 방안의 필요성이 제기되고 있다. 본 연구에서는 인구 대비 교통사고 사망자 비율이 높은 대전시를 대상지역으로 설정하고 보행자 교통사고 데이터를 수집한 후, 기계학습을 통해 최적알고리즘과 심각도 분류의 주요 인자를 도출하였다. 연구의 결과에 따르면, 적용한 9개 알고리즘 중 앙상블 기반의 학습 기법인 AdaBoost (Adaptive Boosting)와 RF (Random Forest)가 최적의 성능을 보여주었다. 이를 기반으로 도출된 대전시 보행자 교통사고 심각도의 주요 인자는 보행자의 연령이 70대 및 20대이거나 사고유형이 횡단사고에 의한 경우로 나타남에 따라 대전시 보행자 사고 저감 대책을 위한 고려요인으로 제안하였다.

• 국제학술발표논문집
• /
• The 8th International Conference on Construction Engineering and Project Management
• /
• pp.463-481
• /
• 2020

Cutter suction dredgers (CSDs) are widely used in various dredging constructions such as channel excavation, wharf construction, and reef construction. During a CSD construction, the main operation is to control the swing speed of cutter to keep the slurry concentration in a proper range. However, the slurry concentration cannot be monitored in real-time, i.e., there is a "time-lag effect" in the log of slurry concentration, making it difficult for operators to make the optimal decision on controlling. Concerning this issue, a solution scheme that using real-time monitored indicators to predict current slurry concentration is proposed in this research. The characteristics of the CSD monitoring data are first studied, and a set of preprocessing methods are presented. Then we put forward the concept of "index class" to select the important indices. Finally, an ensemble learning algorithm is set up to fit the relationship between the slurry concentration and the indices of the index classes. In the experiment, log data over seven days of a practical dredging construction is collected. For comparison, the Deep Neural Network (DNN), Long Short Time Memory (LSTM), Support Vector Machine (SVM), Random Forest (RF), Gradient Boosting Decision Tree (GBDT), and the Bayesian Ridge algorithm are tried. The results show that our method has the best performance with an R² of 0.886 and a mean square error (MSE) of 5.538. This research provides an effective way for real-time predicting the slurry concentration of CSDs and can help to improve the stationarity and production efficiency of dredging construction.

• Korean Journal of Radiology
• /
• 제21권8호
• /
• pp.987-997
• /
• 2020

Objective: Measurement of the liver and spleen volumes has clinical implications. Although computed tomography (CT) volumetry is considered to be the most reliable noninvasive method for liver and spleen volume measurement, it has limited application in clinical practice due to its time-consuming segmentation process. We aimed to develop and validate a deep learning algorithm (DLA) for fully automated liver and spleen segmentation using portal venous phase CT images in various liver conditions. Materials and Methods: A DLA for liver and spleen segmentation was trained using a development dataset of portal venous CT images from 813 patients. Performance of the DLA was evaluated in two separate test datasets: dataset-1 which included 150 CT examinations in patients with various liver conditions (i.e., healthy liver, fatty liver, chronic liver disease, cirrhosis, and post-hepatectomy) and dataset-2 which included 50 pairs of CT examinations performed at ours and other institutions. The performance of the DLA was evaluated using the dice similarity score (DSS) for segmentation and Bland-Altman 95% limits of agreement (LOA) for measurement of the volumetric indices, which was compared with that of ground truth manual segmentation. Results: In test dataset-1, the DLA achieved a mean DSS of 0.973 and 0.974 for liver and spleen segmentation, respectively, with no significant difference in DSS across different liver conditions (p = 0.60 and 0.26 for the liver and spleen, respectively). For the measurement of volumetric indices, the Bland-Altman 95% LOA was -0.17 ± 3.07% for liver volume and -0.56 ± 3.78% for spleen volume. In test dataset-2, DLA performance using CT images obtained at outside institutions and our institution was comparable for liver (DSS, 0.982 vs. 0.983; p = 0.28) and spleen (DSS, 0.969 vs. 0.968; p = 0.41) segmentation. Conclusion: The DLA enabled highly accurate segmentation and volume measurement of the liver and spleen using portal venous phase CT images of patients with various liver conditions.

• 대한전자공학회:학술대회논문집
• /
• 대한전자공학회 2004년도 학술대회지
• /
• pp.671-674
• /
• 2004

We propose accurate classification method of EEG signals during mental tasks. In the experimental task, the tasks of subjects show 3 major measurements; there are mathematical tasks, color decision tasks, and Chinese phrase tasks. The classifier implemented for this work is a feed-forward neural network that trained with the error back-propagation algorithm. The new BCI system is proposed by using neural network. In this system, tr e architecture of the neural network is composed of three layers with a feed-forward network, which implements the error back propagation-learning algorithm. By applying this algorithm to 4 subjects, we achieved $95{\%}$ classification rates. The results for BCI mathematical task experiments show performance better than those of the Chinese phrase tasks. The selection time of each task depends on the mental task of subjects. We expect that the proposed detection method can be a basic technology for brain-computer interface by combining with left/right hand movement or yes/no discrimination methods.

• 말소리와 음성과학
• /
• 제1권2호
• /
• pp.15-22
• /
• 2009

Prominence of speech, which is often called 'accent,' affects the fluency of speaking American English greatly. In this paper, we present an accurate prominence detection method that can be utilized in computer-aided language learning (CALL) systems. We employed pitch movement, overall syllable energy, 300-2200 Hz band energy, syllable duration, and spectral and temporal correlation as features to model the prominence of speech. After the features for vowel syllables of speech were extracted, prominent syllables were classified by SVM (Support Vector Machine). To further improve accuracy, the differences in characteristics of neighboring syllables were added as additional features. We also applied a speech recognizer to extract more precise syllable boundaries. The performance of our prominence detector was measured based on the Intonational Variation in English (IViE) speech corpus. We obtained 84.9% accuracy which is about 10% higher than previous research.

• 한국경영정보학회:학술대회논문집
• /
• 한국경영정보학회 2008년도 춘계학술대회
• /
• pp.446-453
• /
• 2008

All are organizations embedded, here in after, Em-organizaion that confronts the ever-growing complexity. It is important to know Em-organization through Individual Mastery. The complexity must be decreased, and clarified in order to derive to get our ontology from the influence of others. The opportunity to learn in practice is embedded in processes that the community developed. Driving strategic innovation is achieving breakthrough performance throughout the value chain. We used to express complex unit on matrix which includes only the federal statutes because the role of information technology should be a source of competitive advantages each other. Therefore, we got the idea that integrated both kinds of knowledge to create differentiation by ourselves. This practice is situated the learning of Strategic CoP in e-class seminar of our graduate school. We suggest theoretically two things. One is matrix-based decision. Another is creating new context through systems thinking.