Earticle

Recent advancements in Deep Learning (DL) techniques for fire detection have mitigated various ecological, economic, and environmental impacts. However, single models in existing literature often perform poorly due to their inability to capture relevant features with limited data. In this study, we therefore develop an innovative framework for effective fire detection using teacher-student collaborative knowledge distillation. The framework comprises two main components: the teacher model, which leverages a pretrained InceptionV3, and a customized student model designed to inherit the knowledge from the pretrained InceptionV3 to a pruned student model. The proposed network is evaluated and compared against several competitive techniques using two datasets, MAFire and Yar, with different evaluation metrics, offering higher performance and lower computational cost.

Recently, safety inspection of fire outbreak factors is recognized as very important theme because many fires have occurred such as fires in the work process(welding, equipment work, etc) by the use of various ignition sources as well as increase of combustibles construction due to enlargement trend of new construction. However, similar fire accidents occur repeatedly due to the lack of analysis of the correlation between fire occurrence factors(fire cause, risk factor, ignition source) as well as analysis of fire conditions for each construction types. From this point of view, the purpose of this research is to suggest the methodology of fire prevention to contemplate rational plans of safety management targeting 54 accidents in new construction from 2010 to 2023. In order to achieve the purpose of this research, the flashlight process was analyzed for each types of ignition source, and suggest classification system for the fire causes through the case study. The fire causes and risk factor including ignition sources were analyzed for each facilities and construction types of multiple fire. Based on the case study and related law, we suggest the classification structure of fire cases information of new construction and measures for prevention of fire accident targeting facilities and construction types for priority management.

실험실에서 발생하는 화재 사고는 인명과 재산 피해 위험이 커 신속한 탐지와 체계적인 대응이 요구된다. 본 논문에서는 실험실에서 발생할 수 있는 다양한 화재 상황에 효율적으로 대처하기 위하여 4종류(일반, 유류, 전기, 금속) 화재 시나리오를 포함하는 이미지 데이터 셋을 구축하고, 객체 검출(YOLOv5) 기반 화재 특징 추출과 트리 기반 의사결정(Random Forest) 모형을 결합한 이미지 기반 실험실 화재 유형 분류 모델을 제시하고 평가한다.

본 논문에서는 국내 적층형 물류창고의 스프링클러설계 적용을 위한 수용물품 등급분류기준을 제시하였다. 현장조사와 미 국 NFPA, FMDS, 유럽 EN 및 일본의 분류 기준을 비교분석하고 연소특성 실험을 수행하였으며 이를 기준으로 특급, 상급, 중 급 및 경급의 4단계 등급분류(안)을 제시하였다.

본 논문에서는 이미지 학습 및 분류에 가장 많이 활용되고 있는 CNN모델중의 하나인 VGG19알고리즘 을 이용하여 전기화재시 발생하는 용융 흔적을 분류를 방법을 제안하였다. 전기화재의 직접적인 유무를 감지하기 위한 가장 기본적인 근거는 전선의 녹는 정도와 형태에 따라 전기화재 현장에서 발생하는 전선의 용융흔적이다. 제 안된 방법은 VGG19 알고리즘을 활용하여 용융 흔적의 유무를 검출할 수 있도록 변형하여 사용하였으며, 학습에 필요한 데이터는 연구실에서 실제제적하여 사용하였다. 실험결과 제안된 알고리즘의 최종 검증 정확도는 96.31% 이고, 손실율은 0.1169로 확인되었다. 이와 같이 높은 정확도를 확보한 결과를 통하여 용융흔적 검출 알고리즘이 전기화재 유무 검증에 활용될 수 있는 가능성을 보였다.

In this paper, the VGG19 algorithm was used by applying the transfer learning for the classification of molten traces of electric fire arc-beads data which is one of the most used models in convolutional neural network(CNN) computer vision tasks. The most essential basis for detecting direct indications of electric fires is the melting traces of wires that occur at the site of an electric fire, depending on the severity and shape of the melting. The proposed VGG19 method was altered and used such that it could detect molten traces, and the molten trace data of the wires required for learning were created in the lab. The final validation accuracy result was 96.31% with validation loss of 0.1169. Through the result of securing such high accuracy, the possibility of using the melting trace detection algorithm to verify the presence or absence of an electric fire was shown.

본 연구에서는 우리나라 문화재 화재사례의 관련 자료를 수집하여 원인분류 및 분석하였다. 방화, 실화, 자연적원인, 원인미상, 조사불가 등 5가지로 원인을 분류하였다. 원인분석은 발화열원, 발화요인, 최초착 화물, 발화관련기기, 동력원, 연소확대물, 연소확대 범위, 연소확대 요인 등 8가지로 분석하였다. 대응방안 으로는 첫째, 문화재 안전을 위한 국가적 차원의 안전문화 조성이 시급하다. 둘째, 문화재 화재는 무엇보 다 예방이 우선되어야 한다. 문화재 자체의 패시브(passive)적 방어설비의 기술개발이 필요하다. 셋째, 문 화재 안전에 관한 안전문화가 잘 구축되고 예방이 잘 되더라도, 일단 화재가 발생하면 초기진화가 가장 중요하다. 따라서 개별 문화재에 맞는 문제해결형 R&D 연구개발에 관한 전략체계의 수립이 필요하다.

In this study, we analyzed the causes by collection relevant data of Korean cultural property fire cases. The causes were classified into five categories: arson, misfire, natural causes, unknown causes, and no investigation. The cause analysis was carried out in eight ways: ignition heat source, ignition factor, ignition complex, ignition related equipment, power source, combustion enlargement, combustion expansion range, combustion expansion factor. To the countermeasures, first, it is need to create a national safety culture for the safety of cultural properties. Second, prevention of cultural property fire should be a priority. It needs to develop technologies for passive defense facilities for cultural assets. Third, even if safety culture regarding cultural property safety is well established and well prevented, once a fire occurs, first extinguish is most important. Therefore, it is necessary to establish a strategic system for problem-solving R&D for cultural property.

In this paper, we propose a novel, fast model for detecting fire flame and smoke using object detection and 3D classification, referred to as FastFireDet3D. This model uses NanoDet to quickly identify potential areas representing fire and smoke, followed by a novel 3D classification model based on a spatio-temporal convolutional neural network (STCNN). This two-step process allows for efficient and accurate detection. The average processing time for FastFireDet3D is approximately 40-90ms when run on a CPU, and it achieves an accuracy improvement of 3.45% over traditional Convolutional 3D (C3D) models.

이 연구는 TRIZ기법에 의한 물류창고 화재의 원인분석과 4M을 적용하여 화재예방대책을 제시하였다. 연구결 과는 다음과 같다. 첫째, 창의적 문제해결기법인 TRIZ기법을 적용하여 물류창고 화재원인의 모순을 제시하였다. 둘 째, 인적 요인, 물류창고의 화재대책으로 관리자의 안전 기준, 근로자 안전의식 강화, 샌드위치 패널의 작업자 시공기 술 강화 등을 방안을 분석하였다. 셋째, 기계, 설비적 요인, 물류창고의 화재대책으로 안전시설, 안전장치 확대 설치, 화재 진압장비 도입 및 개발, 샌드위치 패널의 내화성능 향상방안을 제시하였다. 넷째, 작업, 환경적 요인, 물류창고 의 화재대책으로 작업공법에 대한 안전수칙 및 관리감독 강화, 물건 적재 장소에 대한 방화구획설정, 성능위주설계 기반으로 한 방화구획의 설정방안을 제시하였다. 마지막으로, 관리적 요인, 물류창고의 화재예방대책으로 화재 위험도 가 낮은 특정소방대상물, 화재안전기준을 적용하기 어려운 특정소방대상물에 샌드위치 패널이 불연재료 이상 재질 규 정을 검토, 물류냉동 창고에 스프링클러설비를 설치, 샌드위치 패널 구조인 물류창고에는 바닥면적의 크기와 관계없 이 자동설비의 설치를 의무화하되 소급적용하는 방안을 제안한다.

This study analyzed the causes of warehouse fires using a creative problem-solving technique called TRIZ. It identified preventive measures by applying 4M. The results are as follows. First, this study examined the inconsistency among the causes of warehouse fires using TRIZ. Second, it analyzed human factors and fire prevention measures in warehouses such as safety standards for managers, and methods for the promotion of safety consciousness among workers, and for the reinforcement of construction technology for sandwich panel workers. Third, it identified the mechanical and facility factors and fire prevention measures in warehouses such as safety facilities, the expanded installation of safety devices, the adoption and development of fire suppression equipment, and the deployment of methods to improve the fire resistance of sandwich panels. Fourth, it presented working and environmental factors and fire prevention measures in warehouses such as the tightening of safety precautions and the supervision of working methods, and setting fire partitions both in loading places and based on performance-based design. Finally, it proposed managerial factors and fire prevention measures in warehouses such as specific targeting for firefighting with low fire hazards, reviewing the material quality regulations of non-combustible or higher for sandwich panels in the specific target of firefighting that cannot apply fire safety standards, installing sprinklers in cold storage, and mandating the installation of automated facilities with retroactive application regardless of the floor area in the warehouse with a sandwich panel structure.

Detection of fire accelerants from fire residues is critical to determine whether the case was arson or accidental fire. However, to develop a standardized model for determining the presence or absence of fire accelerants was not easy because of high temperature which cause disappearance or combustion of components of fire accelerants. In this study, logistic regression, random forest, and support vector machine models were trained and evaluated from a total of 728 GC-MS analysis data obtained from actual fire residues. Mean classification accuracies of the three models were 63 %, 81 %, and 84 %, respectively, and in particular, mean AU-PR values of the three models were evaluated as 0.68, 0.86, and 0.86, respectively, showing fine performances of random forest and support vector machine models.

This paper aims at the statistical analysis of electrical fire and classification of electrical fire causes to collect electrical fires data efficiently. Electrical fire statistics are produced to monitor the number and characteristics of fires attended by fire fighters, including the causes and effects of fire so that action can be taken to reduce the human and financial cost of fire. Electrical fires make up the majority of fires in Korea(including nearly 30% of total fires according to recent figures), The incorrect and biased knowledge for electrical fires changed the classification of certain types of fires, from non-electrical to electrical. It is convenient and required to develop the standardized form that makes, in the assessment of the cause of electrical fires, the fire fighters directly ticking the appropriate box on the fire report form or making an assessment of a text description. Therefore, it is highly recommended to develop electrical fire cause classification and electrical fire assessment on the fire statistics in order to categorize and assess electrical fires exactly. In this paper newly developed electrical fire cause classification structure, which is well-defined hierarchical structure so that there are not any relationship or overlap between cause categories, is suggested. Also fire statistics systems of foreign countries are introduced and compared.

South Korea is divided into five ecoprovinces and sixteen ecoregions. The criteria for ecoprovince classification are ecosystem connectivity and cultural homogeneity. Ecoregions are classified by cluster analysis. The variables used in the analysis are latitude, longitude, seasonal mean temperature, and seasonal precipitation. The large forest fires occurred in the specific ecoregions including Kangwon coastal ecoregion, WoolYoung coastal ecoregion, Hyungsan Taehwa coastal ecoregion, Upper Nagdong river basin ecoregion and Southeastern inland ecoregion. The largest forest fire in the korean history occurred in Kangwon coastal ecoregion in the year 2000. The fire devastated the forestland over 25,000ha. Korea Forest Service, Ministry of Environment, Province Kangwon and NGO organized an investigation committee for the restoration of the burnt area. The committee suggested restoration principles and also forged a restoration strategy of the Kangwon burnt area.

On April 27,1993, a forest fire occurred in Morito-area, Manba-city, Gunma-prefecture Japan. Under the prevailing strong winds, the fire spread and extended to the largest scale ever in Gunma-prefecture. The author chartered a helicopter on May 5, one week after the fire was extinguished, and took aerial photos of tile damaged area, and investigated the condition. of the fire through field survey and data collection. The burnt area extended. over about 100 hectares, and the damage amounted to about 190 million yen (about two million dollar). The fire occurred at a steep mountainous area and under strong winds, therefore, md and topography strongly facilitated the spreading, It is the purpose of this paper to report a damage investigation of the fire and to develop the forecasting method of forest fires based on the topographical analysis and spreading speed of fire. In the first place, I analyze the topographical structure of the regions which became the bject of this study with some topographical factors, and construct a land form classification ap. Secondly, I decide the dangerous condition of each region in the land form classification map according to the direction of the wind and spreading speed of f'kre. In the present paper, I try to forecast forest fires in Morito area, and the basic results for the forecasting method of forest fires were obtained with the topographical classification system and spreading speed of fire.

After the forest fire, as access is not easy, forest damage degree are determined with Landsat TM image rather than visual inspection. Therefore in this study, damaged areas are extracted with factor analysis and cluster analysis. Second factor analysis was performed for areas suspicious as forest fire damage areas to evaluate accuracy after separating into strong, medium and light forest fire areas.

To carry out missions effectively in battlefield environments, it is essential to rapidly identify enemy threats and respond through precise analysis. Accordingly, technologies that utilize artificial intelligence to identify and classify targets in real time are being actively researched in modern warfare. In this study, we propose a deep learning-based target classification algorithm that simultaneously ensures real-time performance and high classification accuracy. Time-series data are constructed from infrared imagery and augmented to train the model, and the effectiveness of the proposed algorithm is demonstrated through comparative experiments with various CNN- and RNN-based models.

전 세계는 기상이변의 영향으로 산불 등 자연 재해가 끊이지 않고 있으며 이로 인한 사회 안전에 심각한 위협이 되고 있다. 특히 대한민국 동해안 지역은 매년 산불 피해로 인한 막대한 재산 피해가 발생하고 있다. 초기 화재 감지 모델의 필수적인 개발은 훈련을 위한 제한된 이미지 데이터와 관련된 도전을 극복해야 하며, 이는 과적합의 위험을 증가시킨다. 이를 해결하기 위해, 랜덤하게 50% 범위까지 회전, 랜덤하게 20% 범위까지 축소 및 확대, 랜덤하게 50%까지 가로 및 세로 뒤집기를 적용하였다. 성능 평가에서 6층 신경망이 7층 신경망보다 더 우수한 성능을 보였으며, 이는 제한된 데이터셋을 가지고 계층 수를 늘리는 것이 바람직하지 않음을 의미한다. 또한, 화재 이미지 분류를 위한 딥러닝 기반 CNN 모델과 ResNet50 전이 학습 모델의 평가는 전이 학습의 우수한 효과를 확인하였다. 이러한 발견은 초기 화재 감지 모델 개발에 도움이 될 것으로 기대하며, 미래 시스템을 위한 귀중한 통찰력을 제공할 것이다.

Natural disasters, such as wildfires, due to climate change are a constant and serious threat to the world and societal safety. Every year, the eastern coastal region of South Korea experiences significant property damage due to wildfires. The imperative development of early fire detection models necessitates overcoming challenges associated with limited image data for training, which elevates the risk of overfitting. To address this, data augmentation techniques, including random rotation (up to 50%), random scaling (up to 20%), and random horizontal and vertical flipping (up to 50%), were employed to augment the training dataset. Performance evaluation indicated that the 6-layer neural network outperformed its 7-layer counterpart, highlighting the impracticality of increasing layer count with a limited dataset. Furthermore, an assessment of deep learning-based CNN models and ResNet50 transfer learning models for fire image classification underscored the superior efficacy of transfer learning. These findings hold promise for advancing early fire detection model development, offering valuable insights for future systems.

This study focuses on the classification of fire equipments for certification based on the risk evaluation. In general, known statistics on fire equipment-related accidents needs to be used for risk evaluation. When statistics is not available, however, expected frequency and severity of accident for individual equipment can be taken into account in evaluating the related risks. Based on the level of inherent risks, each equipment is then classified into three categories for certification. For equipments that risk evaluation is not possible, characteristics of those products such as reliability are considered for classification. Once classified, each equipment is assigned an appropriate certification module.

Korean city had been rapid developed through high industrialization and rapid economic growth after the 1970's. The city development process was guaranteed the fulfillment of city function through the city expansion based on New Towns and satellite cities rather than the reformation of existing city. This city expansion caused by degrading of city, being backward and producing many problems. To solve this problems, it was considered the rehabilitation of retarded function with rejuvenation such as a special law accelerating urban renaissance and reorganization project. Also a fire protection performance did not satisfied the required conditions of modern FPP system, such as the function deterioration of building facilities, the technological development with FPP facilities, changed fire laws and building occupancy for social needs. Insufficiency of requirements depreciated the building value and intimidated a safety of residents. To solve this problem, the improvement of FPP was essentially required and also strongly recommended an analysis and a comparative study between the required FPP of existing building and it improving effective FPP. Therefore, purpose of this study is that establish the basic data to construct a system to analyze and compare the building FPP.

본 연구에서는 강원도 삼척시 지역을 대상으로 산불발생관련 16개 인자별로 주제도로 구축하여 격자 자료로 변환하고 지역단위 산불발생확률을 산출함으로써 산불발생위험지도를 작성하고 진화자원 및 산불감시시설 배치의 적합성을 분석하였다. 또한, 삼척시 지역단위에 해당하는 산불발생확률 모형 추정식을 산출하여 5단계의 산불발생위험등급 지역을 분류하였으며, 산불감시탑, 감시카메라 및 감시초소의 산불감시 범위는 47.9% 정도의 가시권역을 감시할 수 있는 것으로 판단되었다.

Chinese characters are not merely symbolic representations of the Chinese language but also a systematic reflection of the worldview, cognitive approaches, and socio-cultural concepts of ancient Chinese people. This study employs cognitive linguistics, specifically the prototype theory, along with Chinese character structure studies, to analyze the meanings of 127 xiaozhuan script forms that include the structural element '火' as recorded in Shuowen Jiezi. The aim is to explore the prototype meaning of '火' and its semantic extension patterns, providing insight into how ancient Chinese people perceived and categorized fire. The findings reveal that the structural element '火' in Shuowen Jiezi extends from its prototype meaning of physical "fire" to encompass seven fundamental categories: the physical properties of fire, its functions, actions related to fire, disasters, diseases, tools involving fire, and others. This indicates a multifaceted understanding and application of fire by ancient Chinese people. Through the meaning categorization of Chinese character components, this study demonstrates that a cognitive linguistic approach can offer new perspectives for understanding the perceptions and cultural concepts of ancient Chinese society. Additionally, it lays the groundwork for further research into the cognitive conceptualization of Chinese character meanings, aiming to advance the study of the overall semantic system of Chinese characters through individual component analysis.