The Society of Convergence Knowledge Transactions ISSN:2287-8920(Print) 융복합지식학회논문지

Based on the psychological stability and physiological activity of aroma oil, this study aimed to develop a functional hair care product that alleviates scalp and hair damage and improves overall hair health of modern people exposed to daily stress. In particular, we designed a folded aroma hair tonic for improving the scalp that can help protect and recover damaged hair by chemical procedures such as dyeing, bleaching, and perm, and its efficacy was verified through scientific analysis. As a result of measuring the pre- and post-taste brain waves of the folded aroma tonic, the psychological stability and stress relief effect were clearly shown, and positive emotional changes were also confirmed in all items in the K-POMS evaluation. In addition, in the antibacterial activity test, a clear zone of about 1.8 mm was formed compared to the general tonic, proving excellent antibacterial effect, and in the thermal imaging analysis, an immediate reduction in skin temperature and continuous cooling effect were observed. Finally, through the analysis of scalp pores, it was found that the overall scalp environment was improved by reducing dead skin cells and improving moisturization after using the product.

This study aims to vividly reproduce the monochrome or low-resolution images of major great figures repeatedly covered in Korean history education through artificial intelligence-based high-resolution restoration and colorization technology. Furthermore, it is intended to be implemented as a digital hologram through 3D restoration and multi-view image extraction from 2D images. This technological approach has academic and social value not only in the development of future historical education contents but also in the preservation of digital cultural heritage.The experiment utilizes black-and-white and low-resolution images of 9 great figures in elementary, middle, and high school history education in Korea, and then the entire process of depth estimation, multi-view image synthesis, and hologram encoding is performed.

This study applies a Converted Measurement Kalman Filter (CMKF)–based state estimation method to address performance degradation caused by the nonlinearity and noise inherent in polar-coordinate measurements of 2D radar systems. CMKF compensates for bias and covariance distortion introduced during the conversion of range–azimuth measurements into Cartesian coordinates, providing more stable state estimates. In addition, a Double-Gate structure is proposed, in which a small gate is used during constant-velocity intervals to reduce false associations, while an expanded gate is applied during maneuvering intervals to prevent association failures. Simulation experiments covering various motion scenarios—including straight motion, turning, and rapid acceleration—demonstrate that the proposed method achieves consistent performance in terms of association rate, track continuity, and prediction error. These results confirm that the CMKF with the proposed Double-Gate strategy is an effective approach for stable radar target tracking in highly maneuvering environments.

Since the transition to an industrialized society, public concern regarding air pollutants has increased. Air pollution exerts detrimental effects on both human health and the environment, with major pollutants such as fine particulate matter (PM) and ozone posing significant risks to human well-being. These risks are particularly severe for vulnerable populations, including children and the elderly, for whom exposure can be more fatal. Consequently, reducing the concentration of air pollutants is directly linked to safeguarding public health. In this study, we applied tree-based machine learning algorithms—CatBoost, XGBoost, Random Forest, and LSTM—to forecast ozone concentrations using AirKorea data collected from three districts in Seoul between January 2022 and December 2024. The prediction models incorporated not only ozone but also O₂, CO, NO₂ PM2.5, temperature, solar radiation, as well as derived features reflecting ozone lag characteristics and its 24-hour periodic fluctuations. The results showed that CatBoost achieved superior performance in short-, medium-, and long-term forecasting. These findings support the view that variable selection and model architecture play critical roles in improving prediction accuracy in ozone forecasting. Furthermore, the results can serve as foundational data for developing customized prediction models across different regions and time periods. From a practical perspective, the proposed models can be utilized to establish localized ozone prediction services and early warning systems, thereby enabling proactive responses and contributing to the formulation of tailored air quality management policies.

2D LiDAR-based Hector SLAM often fails to update its pose in symmetric corridor environments due to unstable feature matching caused by repetitive geometric structures. Accordingly, the estimated forward displacement remains unchanged even when the robot continues to move, resulting in a forward displacement degeneracy region. To overcome this limitation, this study proposes a forward displacement correction method using a ToF sensor mounted on a stepper motor. The proposed method rotates the ToF sensor at a fixed angular interval, detects abrupt distance changes at corridor structure edges, extracts the last valid corner distance, and calculates the robot’s forward displacement based on trigonometric method. The calculated forward displacement is reflected in the SLAM coordinate frame to compensate for the pose in the displacement degeneracy region. Experimental results demonstrate that the proposed method effectively restores forward position updates in symmetric corridor environments and significantly reduces localization errors without requiring additional sensors or complex sensor fusion.

The purpose of this study is to develop a machine learning-based team win rate prediction model using Major League Baseball (MLB) team-level performance data and to derive implications for training strategy decision-making through interpretation of prediction results. Season data from 2015 to 2024 were used to establish linear regression and rule-based models as baselines, and Random Forest and XGBoost models were applied to compare predictive performance. The experimental results show that machine learning-based models outperform baseline models overall, with the XGBoost model demonstrating the highest explanatory power and stability. Furthermore, variable importance analysis excluding aggregate performance indicators reveals that pitching-related metrics, such as on-base suppression and run prevention, exhibit higher importance than offensive metrics. This suggests that team win rates can be explained more stably by pitching performance in addition to offensive productivity. The findings indicate that machine learning models can be extended beyond simple prediction tools to data-driven decision support systems for team training strategies and resource allocation.

This study examines the determinants of financial performance of PPs using panel data. Fixed-effects regression, dynamic panel GMM, and rolling regression analyses consistently show that NPM has a positive effect on both ROE and ROA. This suggests that the financial performance of PPs is primarily driven by fundamental operating profitability rather than asset efficiency or financial leverage. In contrast, AT does not exhibit a significant effect, reflecting the characteristics of the broadcasting industry, while an increase in EM negatively affects ROE. In line with structural changes in the broadcasting environment, profitability has simultaneously deteriorated across multiple PPs as of 2024, indicating a paradigm shift from growth-oriented to survival-focused operations. Based on these findings, this study emphasizes the importance of prioritizing content competitiveness, cost efficiency, and revenue diversification for achieving sustainable performance in the PP market.

Videofluoroscopic Swallowing Study (VFSS) is the gold standard for evaluating dysphagia. However, it is limited by delayed interpretation times and subjective variability among raters. To address these issues, recent domestic and international studies have focused on integrating Artificial Intelligence (AI) to enable quantitative and standardized analysis of VFSS images. In this study, we propose a Hierarchical Deep Learning Model designed to automatically classify penetration and aspiration, the primary symptoms observed in VFSS. High-quality VFSS videos with ensured clarity and visibility were selected for training and evaluation. The proposed model features a sequential pipeline architecture consisting of three stages: object detection (YOLOv8), semantic segmentation (U-Net), and image classification (ResNet18). Each module within the pipeline was independently trained and evaluated to verify its performance and practical efficacy. Our results demonstrate the potential of this integrated deep learning framework as a diagnostic support tool for dysphagia. We anticipate that further refinement of the object detection module will lead to a fully automated diagnostic system, significantly enhancing diagnostic efficiency and accuracy in clinical environments.

Recently, Audio-based deep learning applications face significant challenges in acquiring high-quality data due to high collection costs, ethical considerations, and legal constraints. Existing generative models are limited by single-resolution and single-scale architectures, which hinder simultaneous modeling of short and long-term temporal patterns. This paper proposes MRMS-WaveGAN, a waveform-domain data augmentation model integrating a multi-resolution generator and a multi-scale discriminator, built upon WaveGAN, which generates audio directly from raw waveforms without spectrogram conversion to preserve phase information. The multi-resolution generator comprises three parallel branches at low, mid, and high resolutions. The low-resolution branch applies Self-Attention to capture long-range temporal dependencies beyond the receptive field of convolution. The mid-resolution branch uses a Residual Block to preserve macro-structural representations while incorporating intermediate acoustic features. The high-resolution branch generates fine-grained waveform details via stacked transposed convolutions. Outputs from all branches are integrated through a fusion module, producing waveforms that complementarily reflect multi-scale temporal features. The multi-scale discriminator evaluates inputs at original, 2×, and 4× downsampled scales in parallel, providing multi-level feedback from high-frequency artifact detection to long-term structural consistency assessment, with Spectral Normalization applied to ensure training stability.

Recent higher education has increasingly been challenged to move beyond credit-based grading systems toward diagnosing and communicating students’ actual competencies. In policy contexts such as the Glocal University initiative, where educational outcomes and industry alignment are emphasized, the coherence between university-developed competencies and industrial demands has become a critical issue. This study explores the nature of the perceived competency gap between universities (supply) and industry (demand) through a mixed qualitative case study. Focusing on H University, the study analyzes its competency assessment framework and skill-print initiatives alongside industry cases. The findings indicate that some perceived gaps result from differences in evaluation criteria and representational language, while substantive gaps persist in areas such as adaptability to change and the application of emerging technologies. Industry stakeholders expressed demand for digital badges but emphasized that fragmented credentials are insufficient without narrative information capturing performance context and developmental processes. Accordingly, this study proposes a rubric-based skill-print framework that structures and accumulates evaluation outcomes around project-based learning.