Earticle

교통사고로 인한 인체 상해를 최소화할 수 있는 안전벨트의 착용이 전 좌석에서 의무임에 도 불구하고, 안전벨트를 착용하지 않은 교통사고 사상자수는 여전히 많은 비율을 차지한다. 주행차량 내 전 좌석 안전벨트 착용 검지시스템은 탑승자의 안전벨트 착용 여부를 판단하고 착용을 유도하는 시스템으로 안전벨트 미착용 사상자수 감소를 위한 유용한 기술적 대응방안 이 될 수 있다. 본 연구에서는 첫 번째로 순서형프로빗 모형을 활용한 교통사고 상해심각도 영향요인 평가를 통해 시스템 도입에 대한 타당성을 제시하였다. 분석결과, 안전벨트 착용 여 부는 교통사고 상해심각도에 통계적으로 유의한 영향을 미치며 안전벨트 착용 시 사망 및 치 명상에 이를 확률이 0.054배 감소하는 것으로 나타났다. 두 번째로 안전벨트 및 교통사고 상해 와 관련된 선행연구를 기반으로 메타분석을 수행하여 시스템 도입 시 기대되는 사고심각도 감 소효과를 추정하였다. 사고심각도 감소효과 분석결과, 안전벨트 착용 시 사망사고가 63.3% 감 소할 것으로 분석되었으며 앞좌석은 75.7%, 뒷좌석은 58.1%의 사망사고 감소효과를 갖는 것으 로 나타났다. 마지막으로 메타분석 결과와 교통사고 통계자료를 기반으로 시스템 도입 시 기 대되는 교통사고 사상자수 감소효과를 도출하여 시스템의 교통안전 측면 효과를 분석하였다. 시스템 도입률이 증가함에 따라 안전벨트 미착용 교통사고 사상자수가 감소하였으며 시스템 도입률이 60%일 경우 사망자수는 현재 대비 3배 이상 감소할 것으로 예상되었다. 분석결과를 활용하여 안전벨트 착용률을 높이고 사고심각도를 저감시킬 수 있는 시스템 운영 방안을 제시 하였다.

Although it is mandatory to wear a seat belt that can minimize human injury when traffic accident occurs, the number of traffic accident casualties not wearing seat belts still accounts for a significant proportion.The seat belt wearing detection system for all seats is a system that identifies whether all seat passengers wear a seat belt and encourages their usage, also it can be a useful technical countermeasure. Firstly, this study established the viability of system implementation by assessing the factors influencing the severity of injuries in traffic accidents through the development of an ordered probit model. Analysis results showed that the use of seat belts has statistically significant effects on the severity of traffic accidents, reducing the probability of death or serious injury by 0.054 times in the event of a traffic accident. Secondly, a meta-analysis was conducted based on prior research related to seat belts and injuries in traffic accidents to estimate the expected reduction in accident severity upon the implementation of the system.The analysis of the effect of accident severity reduction revealed that wearing seat belts would lead to a 63.3% decrease in fatal accidents, with the front seats showing a reduction of 75.7% and the rear seats showing a reduction of 58.1% in fatal accidents. Lastly, Using the results of the meta-analysis and traffic accident statistics, the expected decrease in the number of traffic accident casualties with the implementation of the system was derived to analyze the traffic safety effects of the proposed detection system. The analysis demonstrated that with an increase in the adoption rate of the system, the number of casualties in accidents where seat belts were not worn decreased. Specifically, at a system adoption rate of 60%, it is anticipated that the number of fatalities would decrease by more than three times compared to the current scenario. Based on the analysis results, operational strategies for the system were proposed to increase seat belt usage rates and reduce accident severity.

본 연구는 생물테러리즘의 발생시 이에 대응하는 개념과 선진국들의 생물테러리즘의 대응태세를 알아본 후 한국의 대응태세를 알아보는 연구이다. 생물테러리즘의 발생에 대응하는 태세는 사건의 발생 흐름에 따라 탐지단계, 보호단계, 진단단계, 제독 및 해독단계로 구분해 볼 수 있다. 탐지단계는 한국은 접촉식으로 개발하고 있으나, 선진국에서는 원거리에서 탐지할 수 있는 장비를 개발하고 있다. 한국도 원거리에서 탐지할 수 있는 체계를 개발해야 할 것이다. 접촉식 장비도 민삼하게 작동할 수 있는 장비로 발전 되어야 한다. 보호장비중 한국의 방독면은 세계적으로 그 품질이 우수하나, 기타 피복류 등에 대해서는 첨단 과학기술을 적용할 수 있도록 개선되어야 한다. 진단 장비도 초동조치팀이 현장에 출동해서 즉시 판단할 수 있도록 하는 장비로 개발되어야 한다. 제독을 위한 물자는 인체에 무해하고, 장비에도 무리를 주지 않는 새로운 물질로 개선되는 추세로서 한국에도 이러한 물자가 확보되어야 한다. 해독을 위한 기술은 백신 및 항생제를 개발하는 것으로서 전 세계적으로 노력을 공유하여 개발할 필요가 있으며, 이를 위해 한국의 의료기술을 더욱 발전시킬 필요가 있다. 이외에도 대응매뉴얼, 훈련 모델, 공보 노력, 트라우마 증후군 대비 등의 노력을 기울여야 할 것이다.

This study examines the concepts of counter-act against the Bio-terrorism and the reaction system in advanced countries, thereby to find out reaction system necessary to Korea. Acts on anti-Bio-terrorism is divided to detection stage, protection stage, diagnosis stage and detoxication and neutralization stage according to flow of event occurrence. As for detection stage, Korea is developing it as contact type, while advanced countries are under development of the devices that may detect the terrorism from the remote distance. It is necessary for Korea to develop the remote-distant detection system as well as the contact type of device that may promptly operate. Among the protection gears, the quality of Korea's gas mask is recognized worldwide, but that of other outfits should be improved by applying the state-of-art science technology. The diagnosis device also should be developed to the extent that the dispatched initial action team may make immediate decisions necessary in the field. As the current trends for detoxication materials worldwide require the improvement to new materials harmless to human body and equipment, Korea is also required to acquire those materials. The technology for neutralization means the development of vaccine and antibiotics and it requires the development made by shared efforts worldwide. For this purpose, it is necessary to further develop Korea's medical technology. In addition, the further efforts are required in terms of reaction manual, training model, public communication efforts and preparation for trauma syndrome.

차량 기술이 성장하면서 운전자의 개입이 필요 없는 자율주행까지 발전하였고, 이에 따라 차량 내부 네트워크인 CAN 보안도 중요해졌다. CAN은 해킹 공격에 취약점을 보이는데, 이러한 공격을 탐지하기 위해 기계학습 기반 IDS가 도입된다. 하지만 기계학습은 높은 정확도에도 불구하고 적대적 예제에 취약한 모습을 보여주었다. 본 논문에서는 IDS를 회피할 수 있도록 feature에 잡음을 추가하고 또한 실제 차량의 physical attack을 위한 feature 선택 및 패킷화를 진행하여 IDS를 회피하고 실제 차량에도 공격할 수 있도록 적대적 CAN frame 생성방법을 제안한다. 모든 feature 변조 실험부터 feature 선택 후 변조 실험, 패킷화 이후 전처리하여 IDS 회피실험을 진행하여 생성한 적대적 CAN frame이 IDS를 얼마나 회피하는지 확인한다.

As vehicle technology has grown, autonomous driving that does not require driver intervention has developed. Accordingly, CAN security, an network of in-vehicles, has also become important. CAN shows vulnerabilities in hacking attacks, and machine learning-based IDS is introduced to detect these attacks. However, despite its high accuracy, machine learning showed vulnerability against adversarial examples. In this paper, we propose a adversarial CAN frame generation method to avoid IDS by adding noise to feature and proceeding with feature selection and re-packet for physical attack of the vehicle. We check how well the adversarial CAN frame avoids IDS through experiments for each case that adversarial CAN frame generated by all feature modulation, modulation after feature selection, preprocessing after re-packet.

This study aimed to develop an advanced anomaly-detection system tailored for solar power distribution panels using thermal imaging cameras to ensure operational stability. It addresses the imperative shift toward digitalized safety management in electrical facilities, transcending the limitations of conventional empirical methodologies. Our proposed system leverages a faster R-CNN-based artificial intelligence model optimized through meticulous hyperparameter tuning to efficiently detect anomalies in distribution panels. Through comprehensive experimentation, we validated the efficacy of the system in accurately identifying anomalies, thereby propelling safety protocols forward during the fourth industrial revolution. This study signifies a significant stride toward fortifying the integrity and resilience of solar power distribution systems, which is pivotal for adapting to emerging technological paradigms and evolving safety standards in the energy sector. These findings offer valuable insights for enhancing the reliability and efficiency of safety management practices and fostering a safer and more sustainable energy landscape.

Broilers are bred at a higher density than large livestock, and their relatively small size makes monitoring challenging, requiring a reliable, cost-effective automatic monitoring system for precision farming. Particularly in the case of managing large-scale broiler breeding systems, such as multi-tier broiler houses, it is important to reduce investment in computing facilities and to improve productivity through automatic monitoring technology. However, financial ability of farmers to equip and manage the high-specification computing equipment required by these technologies is limited, necessitating development of affordable monitoring systems. We there propose a lightweight broiler-detection model that augments YOLOv5 with two efficiency modules: Bi-directional Feature Pyramid Network (BiFPN), which fuses multiscale features with learnable weights to sharpen small-object and GhostNet, which replaces many convolutions with inexpensive “ghost” operations to cut parameters. Performance was benchmarked against the original YOLOv5 on a high-spec local GPU machine and CPU based cloud platform. The proposed model attained 90.5% mAP@0.5, marginally below the baseline’s 92.8%, yet inference time dropped by 30% on CPU (50.9 ms) and 50% on GPU (1.8 ms). By implementing lightweight models, farmers can utilize monitoring systems from anywhere via mobile devices, leveraging cloud-based technology. This eliminates the necessity for expensive hardware, offering an affordable and practical solution for improving farming management.

Pedestrian tracking is a particular object tracking problem and an important component in various vision-based applications, such as autonomous cars and surveillance systems. Following several years of development, pedestrian tracking in videos remains challenging, owing to the diversity of object appearances and surrounding environments. In this research, we proposed a tracking-by-detection system for pedestrian tracking, which incorporates a convolutional neural network (CNN) and color information. Pedestrians in video frames are localized using a CNN-based algorithm, and then detected pedestrians are assigned to their corresponding tracklets based on similarities between color distributions. The experimental results show that our system is able to overcome various difficulties to produce highly accurate tracking results.

금융분야의 파괴적 혁신인 디지털 전환은 금융업을 근본적으로 변화시키고 금융소비자에게 새로운 경험과 기회를 제공하 고 있다. 그러나 이와 동시에 보이스피싱 등 디지털금융사기 역시 IT와 결합하여 점차 고도화하고 있다. 각 금융회사에서는 디지털금융사기에 대응하고 금융소비자의 피해를 방지하기 위해 FDS(Fraud Detection System)를 구축·운영하고 있으나 관련 연구 대부분이 AI 등을 활용한 기술적인 연구에 집중되어 있으며, FDS 수준이 낮거나 신규구축하고자 하는 금융업권별 FDS 구축·운영 방법론에 대한 실증적인 연구는 현재까지 부족한 상황이다. 본 논문은 이러한 문제점 해소를 위해 각 금융업권별 비대면업무, 발생하는 디지털금융사기 유형 및 가용 데이터 등에 대한 분석 후 이를 FDS 업무 프로세스와 결합하여 FDS 프 레임워크 및 특성 변수를 마련하고 이를 바탕으로 각 금융업권별 FDS 구축·운영 방법론을 제시한다.

Digital transformation is fundamentally reshaping the financial industry, offering consumers new experiences and opportunities. However, alongside this, digital financial fraud such as voice phishing is becoming increasingly sophisticated by integrating with IT. Financial institutions implement Fraud Detection Systems (FDS) to combat these crimes and protect consumers, yet most research about FDS focuses on technological studies such as AI-based FDS. Practical methodologies for developing and operating FDS across various financial sectors remain scarce, especially for institutions with underdeveloped FDSs. This paper addresses these challenges by analyzing non-face-to-face operations, digital financial fraud types, and available data for each sector, integrating these with FDS processes. It proposes frameworks and variables for FDS development tailored to each financial sector.

This paper introduces an edge AI-based scene-specific object detection system for long-term traffic management, focusing on analyzing congestion and movement via cameras. It aims to balance fast processing and accuracy in traffic flow data analysis using edge computing. We adapt the YOLOv5 model, with four heads, to a scene-specific model that utilizes the fixed camera's scene-specific properties. This model selectively detects objects based on scale by blocking nodes, ensuring only objects of certain sizes are identified. A decision module then selects the most suitable object detector for each scene, enhancing inference speed without significant accuracy loss, as demonstrated in our experiments.

This study proposes a novel video traffic flow detection method based on machine vision technology. The three-frame difference method, which is one kind of a motion evaluation method, is used to establish initial background image, and then a statistical scoring strategy is chosen to update background image in real time. Finally, the background difference method is used for detecting the moving objects. Meanwhile, a simple but effective shadow elimination method is introduced to improve the accuracy of the detection for moving objects. Furthermore, the study also proposes a vehicle matching and tracking strategy by combining characteristics, such as vehicle's location information, color information and fractal dimension information. Experimental results show that this detection method could quickly and effectively detect various traffic flow parameters, laying a solid foundation for enhancing the degree of automation for traffic management.

The most important things for a forest fire detection system are the exact extraction of the smoke from image and being able to clearly distinguish the smoke from those with similar qualities, such as clouds and fog. This research presents an intelligent forest fire detection algorithm via image processing by using the Gaussian Mixture model (GMM), which can be applied to detect smoke at the earliest time possible in a forest. GMMs are usually addressed by making the model adaptive so that its parameters can track changing illuminations and by making the model more complex so that it can represent multimodal backgrounds more accurately for smoke plume segmentation in the forest. Also, in this paper, we suggest a way to classify the smoke plumes via a feature extraction using HSL(Hue, Saturation and Lightness or Luminanace) color space analysis.

The Internet explosion and the increase in crucial web applications such as ebanking and e-commerce, make essential the need for network security tools. One of such tools is an Intrusion detection system which can be classified based on detection approachs as being signature-based or anomaly-based. Even though intrusion detection systems are well defined, their cooperation with each other to detect attacks needs to be addressed. Consequently, a new architecture that allows them to cooperate in detecting attacks is proposed. The architecture uses Software Agents to provide scalability and distributability. It works in two modes: learning and detection. During learning mode, it generates a profile for each individual system using a fuzzy data mining algorithm. During detection mode, each system uses the FuzzyJess to match network traffic against its profile. The architecture was tested against a standard data set produced by MIT's Lincoln Laboratory and the primary results show its efficiency and capability to detect attacks. Finally, two new methods, the memory-window and memoryless-window, were developed for extracting useful parameters from raw packets. The parameters are used as detection metrics.

Recently, there has been an increasing demand for computer-vision-based inspection and/or measurement system as a part of factory automation equipment. In general, it is almost impossible to check the fault of all parts, coming from part-feeding system, with only manual inspection because of time limitation. Therefore, most of manual inspection is applied to specific samples, not all coming parts, and manual inspection neither guarantee consistent measuring accuracy nor decrease working time. Thus, in order to improve the measuring speed and accuracy of the inspection, a computer-aided measuring and analysis method is highly needed. In this paper, a computer-vision-based pipe inspection system is proposed, where the front and side-view profiles of three different kinds of pipes, coming from a forming line, are acquired by computer vision. And the edge detection is processed by using Laplace operator. To reduce the vision processing time, modified Hough transform is used with clustering method for straight line detection. And the center points and diameters of inner and outer circle are found to determine eccentricity of the parts. Also, an inspection system has been built so that the data and images of faulted parts are stored as files and transferred to the server.

Computer security has become a critical issue with the rapid development of business and other transaction systems over the Internet. The application of artificial intelligence, machine learning and data mining techniques to intrusion detection systems has been increasing recently. But most research is focused on improving the classification performance of a classifier. Selecting important features from input data leads to simplification of the problem, and faster and more accurate detection rates. Thus selecting important features is an important issue in intrusion detection. Another issue in intrusion detection is that most of the intrusion detection systems are performed by off-line and it is not a suitable method for a real-time intrusion detection system. In this paper, we develop the real-time intrusion detection system, which combines an on-line feature extraction method with the Least Squares Support Vector Machine classifier. Applying the proposed system to KDD CUP 99 data, experimental results show that it has a remarkable feature extraction and classification performance compared to existing off-line intrusion detection systems.

With the increasing sophistication of cyber threats, traditional Intrusion Detection Systems (IDS) often fail to adapt to evolving attack patterns, leading to high false positive rates and inadequate detection of zero-day attacks. This study proposes the Deep Q-Learning Intrusion Detection System (DQ-IDS), a novel reinforcement learning (RL)-based approach designed to dynamically learn network attack behaviors and continuously enhance detection performance. Unlike conventional machine learning (ML) and deep learning (DL)-based IDS models that depend on static, pre-trained classifiers, DQ-IDS employs Deep Q-Networks (DQN) with experience replay and adaptive ε-greedy exploration to autonomously classify benign and malicious network traffic. The integration of experience replay mitigates catastrophic forgetting, while adaptive exploration ensures an optimal trade-off between learning efficiency and threat detection. A reward-driven training mechanism reinforces correct classifications and penalizes errors, thereby reducing both false positive and false negative rates. Extensive empirical evaluations on real-world network datasets demonstrate that DQ-IDS achieves a detection accuracy of 97.18%, significantly outperforming conventional IDS solutions in both attack detection and computational efficiency. This work introduces a paradigm shift toward adaptive, self-learning cybersecurity systems capable of real-time, robust threat mitigation in dynamic network environments.

Internet of Things (IoT) is suffered from different types of attacks due to vulnerability present in devices. Due to many IoT network traffic features, the machine learning models take time to detect attacks. This paper proposes a feature selection for intrusion detection systems (IDSs) using Information Gain (IG) and Gain Ratio (GR) with the ranked top 50% features for the detection of DoS and DDoS attacks. The proposed system obtains feature subsets using insertion and union operations on subsets obtained by the ranked top 50% IG and GR features. The proposed method is evaluated and validated on IoT-BoT and KDD Cup 1999 datasets, respectively, with a JRipclassifier. The system provides higher performance than the original feature set and traditional IDSs on IoT-BoT and KDD Cup 1999 datasets using 16 and 19 features, respectively.

Remote non-contact monitoring of human vital signs has recently received lots of attention due to safety requirements. Among a variety of sensors available, millimeter-wave radars show great advancement and advantages. With the miniaturization of radar systems, frequency modulated continuous wave (FMCW) millimeter-wave radar has been employed to address the issue of in-vehicle life detection. In the designed in-vehicle life detection system, Yosemite 4T8R 77/79 GHz radar chip communicates with the engine control unit (ECU). The system can detect the availability of child in car, alarm and perform self-check. It is shown that the designed system has high accuracy and anti-interference ability. It has convenient interface with smart cockpit system and is suitable for a wide range of vehicles.

People often create fake social media accounts to express themselves anonymously. However, these fake accounts can harm the reputation of individuals and businesses, resulting in fewer genuine likes and followers. Instagram, a top-rated social media platform often used for business and political engagement, suffers from the negative impacts of these accounts. This highlights the urgent need for a dependable system to identify whether Instagram accounts are genuine. This study investigated several machine learning models for developing a fake account detection system. Single models, such as support vector machines, naïve Bayes, logistic regression, multilayer perceptron, and ensemble models based on bootstrap aggregating techniques and boosting, were trained and tested. The training and testing processes were conducted using a 10-fold cross-validation to prevent overfitting. The test results indicated that the adaptive and gradient boosting models achieved the best accuracy and an F1 score of more than 92%, with precision surpassing 93%.

Nowadays, the Internet of Things (IoT) network, is increasingly becoming a ubiquitous connectivity between different advanced applications such as smart cities, smart homes, smart grids, and many others. The emerging network of smart devices and objects enables people to make smart decisions through machine to machine (M2M) communication. Most real-world security and IoT-related challenges are vulnerable to various attacks that pose numerous security and privacy challenges. Therefore, IoT offers efficient and effective solutions. intrusion detection system (IDS) is a solution to address security and privacy challenges with detecting different IoT attacks. To develop an attack detection and a stable network, this paper's main objective is to provide a comprehensive overview of existing intrusion detections system for IoT environment, cyber-security threats challenges, and transparent problems and concerns are analyzed and discussed. In this paper, we propose software-defined IDS based distributed cloud architecture, that provides a secure IoT environment. Experimental evaluation of proposed architecture shows that it has better detection and accuracy than traditional methods.

As machine learning technology advances, data security becomes increasingly important. In this study, we propose an intrusion detection mechanism based on federated learning (FL) that updates only the learning weights to minimize the risk of information leakage. Considering the limited resources of industrial Internet of Things (IIoT) nodes, we propose a learning method based on data pruning. The proposed FL-based intrusion detection model was found to be more secure than the centralized model in terms of the data leakage rate. Data pruning technology reduced the memory usage by 1.4 times while maintaining 97.7 % accuracy. The proposed method detects attacks in industrial sites where large-scale IIoT nodes are installed efficiently, and protects industrial secrets and personal information effectively.

We propose a detection algorithm based on tree-structured convolutional neural networks (TsCNNs) that finds pornography, propaganda, or other inappropriate content on a social media network. The algorithm sequentially applies the typical convolutional neural network (CNN) algorithm in a tree-like structure to minimize classification errors in similar classes, and thus improves accuracy. We implemented the detection system and conducted experiments on a data set comprised of 6 ordinary classes and 11 inappropriate classes collected from the Korean military social network. Each model of the proposed algorithm was trained, and the performance was then evaluated according to the images and videos identified. Experimental results with 20,005 new images showed that the overall accuracy in image identification achieved a high-performance level of 99.51%, and the effectiveness of the algorithm reduced identification errors by the typical CNN algorithm by 64.87 %. By reducing false alarms in video identification from the domain, the TsCNNs achieved optimal performance of 98.11% when using 10 minutes frame-sampling intervals. This indicates that classification through proper sampling contributes to the reduction of computational burden and false alarms.