This study presents a truck classification method using panoramic side-view images to meet the Ministry of Land, Infrastructure and Transport’s 12-category standard (types 4–12). The system captures a vehicle’s full side profile via a panoramic imaging device, ensuring complete wheel visibility. A YOLOv12-based deep learning model detects wheels, and image processing extracts their center coordinates. Pixel distances between adjacent wheels are calculated and normalized to determine axle spacing patterns, which, together with wheel count, are applied to a rule-based classifier. Tests on 1,200 real-world panoramic truck images (1,000 for training, 200 for testing) achieved a mean average precision of 96.1% for wheel detection and 90.5% overall classification accuracy. The method offers explainable classification through measurable structural features, supporting applications in smart tolling, road usage billing, overloading enforcement, and autonomous vehicle perception.

This paper reviews ordinal decision tree algorithms for ordinal classification, exploring theoretical foundations, key algorithms (MDT, QMDT), specialized splitting criteria (Ordinal Gini, Weighted Information Gain), and ensemble methods. It discusses applications in healthcare and social sciences, highlighting interpretability and flexibility while acknowledging overfitting and instability. As implications for future research, this study points out advantages such as interpretability and flexibility, and limitations such as overfitting and instability.

Passive acoustic monitoring (PAM) has emerged as an effective tool for studying underwater soundscapes and monitoring marine organisms. In this study, the biological sounds of three fish species that mainly inhabit or occur in the Korean coastal oceans, brown croaker (Miichthys miiuy), Pacific cod (Gadus macrocephalus), and small yellow croaker (Larimichthys polyactis) were recorded using the PAM method. The possibility of automatic classification was evaluated using a deep learning-based convolutional neural network (CNN) model based on the measured data. The biological fish sounds were recorded using hydrophones in the sea cage environments. The three fish species data were converted into spectrogram images and used as input for training and evaluating the CNN model. Gaussian noise was added to the test data to simulate low signal-to-noise ratio (SNR) environments. The model achieved high classification performance, with F1-score of about 96% on raw data and about 77% accuracy under signal-to-noise ratio conditions. These results suggest that CNN-based models be adequate for fish sound classification, even in acoustically complex underwater environments. Applying CNN models to classify and detect fish sounds can improve the automation and efficiency of PAM-based acoustic analysis, thereby improving the monitoring of fish populations, resource assessment, and ecological management in the future.

The casting manufacturing process of aluminum automotive wheels often involves processing various wheel models during stages such as flow forming, machining, packaging, and delivery. Traditionally, separate equipment or production lines were required for each model, which led to higher facility investment costs and increased labor costs for classification. However, the implementation of machine learning-based model classification technology has made it possible to automatically and accurately distinguish between different wheel models, resulting in significant cost savings and enhanced production efficiency. Additionally, this approach helps prevent product mix-ups during the final inspection process and allows for the quick and precise identification of wheel models during packaging and delivery, reducing shipping errors and improving customer satisfaction. Despite these benefits, the high cost of machine learning equipment presents a challenge for small and medium-sized enterprises(SMEs) to adopt such technologies. Therefore, this paper analyzes the characteristics of existing machine learning architectures applicable to the automotive wheel manufacturing process and proposes a custom CNN(Convolutional Neural Network) that can be used efficiently and cost-effectively.

This study aimed to categorize grammar items in English textbooks used across Korean elementary, middle, and high schools by specific grade levels. To achieve this, we developed an automated grammar item analyzer using natural language processing techniques, which analyzed 52,964 sentences from the textbooks. We selected 173 grammar items from the 2022 revised national curriculum and classified them according to the methodology used to determine CEFR levels in the English Grammar Profile. The classification results are as follows: 18 items for Grade 3 of elementary school, 6 for Grade 4, 11 for Grade 5, 8 for Grade 6, 29 for the first year of middle school, 43 for the second year, 22 for the third year, 29 for the first year of high school, and 7 for the second year. Based on these findings, this study discusses pedagogical implications, focusing on practical applications such as refining assessment tools, more precisely defining curricular objectives, and developing grade-specific instructional materials.

This study aims to identify latent classes among shared e-scooter users based on their characteristics and analyze the differences in personal and usage characteristics across these classes. Specifically, the study has the following key objectives: (1) to select variables related to the personal and usage characteristics of shared e-scooter users; (2) to collect data on the personal and usage characteristics of shared e-scooter users; (3) to derive the latent classes of shared e-scooter users; and (4) to test the differences in personal and usage characteristics across the identified latent classes. Variables related to the personal and usage characteristics of shared e-scooter users were selected based on a literature review. Through a survey, data on the personal and usage characteristics of shared e-scooter users were collected. A latent class analysis (LCA) was performed to derive the latent classes of shared e-scooter users. Finally, a chi-square analysis was conducted to test the differences in personal and usage characteristics across the latent classes of shared e-scooter users. The results of this study are as follows. The personal characteristics of shared e-scooter users were identified as age and sex, whereas the usage characteristics were identified as usage frequency, time periods of e-scooter usage, return/rental zones, return/rental places, and types of roads used. Data on sex, age, usage frequency, periods of e-scooter usage, and return/rental locations were collected from 278 shared e-scooter users. Based on information criterion, statistical validation, and the entropy index, four latent classes of shared e-scooter users were identified: “male users with a commuting purpose in business zones,” “male users with a homeward commuting purpose in residential zones,” “female users with a leisure purpose in park/green zones,” and “users in their 20s with a commuting purpose in residential zones.” The results of a chisquare analysis revealed statistically significant differences (p < 0.05) in the personal and usage characteristics across the latent classes. Shared e-scooter user types were classified through Latent Class Analysis (LCA), and differences in personal and usage characteristics were identified across the classes. The preferred usage environments and conditions for each class of shared e-scooter users are determined. Variables related to the return/rental zone and periods of e-scooter usage showed the most significant differences among the classes. These findings can contribute to the development of customized user policies and the improvement of services based on the characteristics of shared e-scooter users.

Crop diseases seriously affect food security, and traditional identification methods are inefficient and inaccurate. This paper proposes a GoogLeNet model with an attention mechanism. By integrating an attention module inside the Inception module, the recognition ability of subtle disease features and complex backgrounds is improved. Based on strict data preprocessing and enhancement, the proposed method achieves 87.75% accuracy on the AI Challenger 2018 crop disease dataset, which is better than the existing advanced methods, which verifies the effectiveness and practicability of the method and provides technical support for smart agriculture.

The Japanese government's longstanding policy on Taiwan and the continuous promotio n of Japanese education, coupled with the economic and cultural exchanges with Japan aft er the restoration of Taiwan, have resulted in a considerable amount of Japanese vocabula ry remaining in Taiwan, which has first entered the Taiwanese vocabulary system.However, as far as these words are concerned, instead of focusing on the writing of words, it is bet ter to present them in the form of sounds, which are naturally present in the language of daily life. These words have not disappeared with the breakup of Taiwan from Japanese ru le, and they are still used regularly, especially for native speakers (73.3% of the population of Taiwan speaks Taiwanese), and the use of Taiwanese interspersed with Japanese vocabul ary is a natural thing to do. Chinese language borrows words from Taiwanese, not only be cause of the lack of Taiwanese culture and daily life words, but also because of the lack o f Chinese vocabulary, as there are still words with similar meanings or synonyms, but to e xpress the lively and vivid character of the Chinese language. In any case, with the freque nt exchanges between Taiwan and Japan, and the increasing use of Chinese in the Taiwan ese society, many Japanese words originally belonging to Taiwanese usage are gradually b eing used as Chinese, with Chinese pronunciations. These loanwords from Japanese, whether they still exist in Taiwanese, or have been intro duced into Chinese from Taiwanese, or have even been introduced directly into the Chines e language system in recent years, can all indicate that the Taiwanese language is a very i mportant part of the Chinese language system. These loanwords from Japanese, whether t hey still exist in Taiwanese, or have been imported from Taiwanese into Chinese, or have even been directly introduced into the Chinese language system in recent years, all illustrat e the fact that Taiwanese culture has been deeply influenced by Japanese culture. The reas on why a large number of Japanese loanwords can be commonly used in daily life is due to the inclusiveness of other cultures and the acceptance of words, and the use of many of these loanwords will be partially changed in the process of cultural exchange, whether i t is an increase in meaning, a change in meaning, or an increase in the number of words in a particular language. Many borrowed words are more or less partially changed in the process of cultural exchanges, no matter whether the meaning of the word is increased, d ecreased or modified, but they can still be used, which shows that the use of language is very flexible. From the exchange of vocabulary, we can also see the sense of identification with the culture of a certain country. In the case of Taiwan, there has always been a deep affection for Japan, and this affection has not diminished over time, as evidenced by the l arge number of loanwords used in daily life in recent years

결빙(Black Ice)은 도로 포장체 표면의 균열 등에 스며든 습기나 눈, 그리고 차량 주행 중 발생하는 타이어 분진 및 배 기가스 등의 영향으로 인해 도로 표면과 유사한 색상의 얇은 얼음막이 형성되는 현상을 의미한다(Cho et al., 2021). 도로 노면이 결빙 상태일 경우, 평균 미끄럼 저항 계수는 건조 노면의 약 30% 수준으로 크게 낮아진다(Lee et al., 2024). 또 한, 결빙은 도로 표면과 색상이 유사하여 운전자가 노면 상태를 즉각적으로 인지하기 어렵고, 이에 따라 제동이나 회피 를 위한 충분한 시간을 확보하기 어렵다. 최근 5년간 발생한 서리·결빙 노면 교통사고의 치사율(사고 100건당 사망자 수) 은 2.69명으로, 이는 건조 노면 교통사고 치사율의 약 2배, 습윤 노면의 1.3배 수준에 해당한다(KoROAD, 2024). 이러한 위험성을 고려하여 국토교통부는 2020년 전국 고속국도 및 일반, 위임국도를 대상으로 403개 구간을 결빙 취약 구간으로 지정하였으며, 이후 464개소로 확대하여 자동염수분사시설, 그루빙(Grovving), 결빙주의표지판 등 안전시설을 확충하여 결빙사고를 집중적으로 관리하고 있다(MOLIT, 2020; BAI 2021). 하지만, 결빙사고 발생건수는 2020년 524건, 2021년 1,204건, 2022년 1,042건으로 증가추세를 보이고 있어, 결빙 취약 구간의 평가 적절성과 실효성에 대한 검토 필요성이 대 두되고 있다(KoROAD, 2024). 본 연구에서는 최근 10년 고속국도에서 발생한 결빙사고와 결빙사고 영향인자를 Random Forest Algorithm으로 분석하 여 도로 구간별 결빙사고 위험도를 평가하였다. 국가교통정보센터의 노드·링크(Node·Link) 체계를 기반으로 전국 고속국 도의 동절기 기상, 기하구조, 교통량 등 결빙사고 영향인자를 구간별로 수집하였다. 각 구간은 최근 10년 결빙사고 데이 터를 통해 결빙사고 발생구간과 비발생 구간으로 분류하였다. 구간별 수집한 결빙사고 영향인자를 독립변수, 사고발생유 무를 종속변수로하여 알고리즘 학습을 위한 데이터셋(Data Set)을 구성하고, 데이터불균형 문제를 해결하기 위해 오버샘 플링(OverSampling) 기법 중 하나인 SMOTE(Synthetic Minority Oversampling Technique)을 적용하였다. 최종적으로 Random Forest Classification Model을 학습하고, 모델의 하이퍼파라미터 조정(HyperParameter Tunning)을 거처 결빙사 고 발생구간 예측성능이 가장 높은 모델을 결정하였다. 이를 통해, 전국 고속국도의 구간별 결빙사고 발생 위험도를 평 가하고 각 결빙사고 영향인자의 변수중요도를 분석함으로써 결빙 취약구간 평가 방안의 신뢰성 제고를 기대한다.

본 연구는 자율주행자동차(AV)와 비자율주행자동차(HDV)가 혼재하는 교통환경에서 자율주행 전용차로 도입 시 차로변경구간 길이가 교통소통지표에 미치는 영향을 VISSIM 시뮬레이션을 통해 분석하였다. 분석을 위하여 도로구조 및 교통량, AV All-knowing, Cautious 주행행태를 반영하여 432개 시나리오를 구성하였다. 시뮬레이션 결과, 차로변경구간 길이가 증가할수록 교통 밀도와 지체시간은 유의 미하게 감소하였으며, 속도 및 통과교통량이 증가하는 효과를 보였으나, 자율주행 전용차로 도입 시 밀도와 지체시간이 증가하고 속도 및 통과교통량이 감소하는 등 일부 부정적 영향을 확인할 수 있었다. 향후 실제 도로 데이터를 반영한 분석을 통해 연구 신뢰성을 제 고할 필요성이 존재한다.

인공지능의 발전은 검색엔진, SNS, ChatGPT 등 다양한 분야에서 혁신을 이끌며 사회와 산업 전반에 변화를 가져오고 있다. 특히, 교 통 분야에서는 AI 기반 기술이 교통정보 수집 및 분석 방식에 변화를 주며, 새로운 활용 가능성을 제시하고 있다. 과거 육안 계수 방 식에 의존했던 교통량 조사는 현재 CCTV 영상과 딥러닝 객체 인식 기술을 활용해 신뢰성과 정확성이 크게 향상되었다. AI 기반 교통 솔루션의 도입으로 교통량 조사 데이터는 정책 수립, 운영 개선, 사회간접자본 건설 등 다양한 분야에서 중요한 기초 자료로 활용되고 있다. 이에 본 연구에서는 YOLO v8을 활용하여 차량 축 인식 기반 차종 분류의 정확성을 향상시키고, 기존 촬영 기법과 비교·분석을 통해 최적의 인식기법을 제시하고자 한다.

선박에는 단열을 위한 발포제가 적용된다. 기존의 발포제에는 지구온난화물질인 수소불화탄소(HFC)를 다량 포함하고 있는 문제점이 있으며, 우리나라는 몬트리올 의정서의 ‘키칼리 개정서’를 채택함에 따라 HFC를 ‘24년부터 ’45년까지 기준 수량의 80% 감 축하기로 결정되었다. 이에, 메틸포메이트 원료는 지구온난화지수가 0(HFC는 960~1,430)으로 향후 친환경발포제로 높은 기대를 갖고 있다. 하지만, 메틸포메이트 발포제의 성능은 원료의 순도 및 주변환경에 높은 영향을 받음으로 각 공정환경에 대한 정확한 분류가 필요하다. 이에, 본 논문에서는 주변환경(온도)과 메틸포메이트 순도에 따라, 총 4개의 케이스를 만들었다. 각 케이스에 대해서 10,010 장의 이미지를 학습하고, 이를 구글넷(GoogLeNet)알고리즘을 이용하여 분류하였다. 분류결과 정확도는 96.8%를 갖고, F1-Score는 0.969 를 갖는 것으로 계산하였다.