Search Results (31)

Urban documents like city planning reports and environmental data often feature complex charts and texts that require effective summarization tools, particularly in smart city management systems. These documents increasingly use graphical abstracts alongside textual summaries to enhance readability, making automated abstract generation crucial. This study explores the application of summarization technology using scientific paper abstract generation as a case. The challenge lies in processing the longer multimodal content typical in research papers. To address this, a deep multimodal-interactive network is proposed for accurate document summarization. This model enhances structural information from both images and text, using a combination module to learn the correlation between them. The integrated model aids both summary generation and significant image selection. For the evaluation, a dataset is created that encompasses both textual and visual components along with structural information, such as the coordinates of the text and the layout of the images. While primarily focused on abstract generation and image selection, the model also supports text–image cross-modal retrieval. Experimental results on the proprietary dataset demonstrate that the proposed method substantially outperforms both extractive and abstractive baselines. In particular, it achieves a Rouge-1 score of 46.55, a Rouge-2 score of 16.13, and a Rouge-L score of 24.95, improving over the best comparison abstractive model (Pegasus: Rouge-1 43.63, Rouge-2 14.62, Rouge-L 24.46) by approximately 2.9, 1.5, and 0.5 points, respectively. Even against strong extractive methods like TextRank (Rouge-1 30.93) and LexRank (Rouge-1 29.63), our approach shows gains of over 15 points in Rouge-1, underlining its effectiveness in capturing both textual and visual semantics. These results suggest significant potential for smart city applications—such as accident scene documentation and automated environmental monitoring summaries—where rapid, accurate processing of urban multimodal data is essential. Full article

As terrestrial resources become increasingly scarce, the exploration and utilization of marine resources have become crucial for ensuring a stable resource supply. A maritime A-Frame is a specialized lifting mechanism mounted on the stern of a vessel, designed for deploying and retrieving heavy loads during subsea exploration. Real-time monitoring of the stress of A-Frames is essential for identifying potential failures and preventing accidents. This paper presents a stress-monitoring campaign conducted on a maritime A-Frame during a deep-sea mining project in the South China Sea. Fiber Bragg Grating (FBG) strain sensors were installed on the A-Frame to measure its stress responses throughout the deep-sea mining operation. The stress variations observed during the deployment and retrieval of a deep-sea mining vehicle were analyzed. The results indicate that the stress caused by the swinging motion of the A-Frame was significantly higher than that generated by the lifting and deployment of the mining equipment. Additionally, a finite element model (FEM) of the A-Frame was developed to estimate the stress of the hot spots by integrating the measured strain data. The analysis confirmed that the maximum stress experienced by the A-Frame was well below the allowable threshold, indicating that the structure had sufficient strength to withstand operational loads. In addition, the swing angle of the A-Frame significantly affects the stress value of the A-Frame, while lifting the mining vehicle has a very slight effect. Thus, it is advisable to accelerate the deployment and retrieval speeds of the mining vehicle and minimize the outward swing angle of the A-Frame. These findings provide valuable insights for optimizing the design and ensuring the safe operation of maritime A-Frames in deep-sea mining exploration. Full article

The timely identification of probable causes in aviation incidents is crucial for averting future tragedies and safeguarding passengers. Typically, investigators rely on flight data recorders; however, delays in data retrieval or damage to the devices can impede progress. In such instances, experts resort to supplementary sources like eyewitness testimonies and radar data to construct analytical narratives. Delays in this process have tangible consequences, as evidenced by the Boeing 737 MAX accidents involving Lion Air and Ethiopian Airlines, where the same design flaw resulted in catastrophic outcomes. To streamline investigations, scholars advocate for natural language processing (NLP) and topic modelling methodologies, which organize pertinent aviation terms for rapid analysis. However, existing techniques lack a direct mechanism for deducing probable causes. To bridge this gap, this study trains and evaluates the performance of a transformer-based model in predicting the likely causes of aviation incidents based on long-input raw text analysis narratives. Unlike traditional models that classify incidents into predefined categories such as human error, weather conditions, or maintenance issues, the trained model infers and generates the likely cause in a human-like narrative, providing a more interpretable and contextually rich explanation. By training the model on comprehensive aviation incident investigation reports like those from the National Transportation Safety Board (NTSB), the proposed approach exhibits promising performance across key evaluation metrics, including BERTScore with Precision: (M = 0.749, SD = 0.109), Recall: (M = 0.772, SD = 0.101), F1-score: (M = 0.758, SD = 0.097), Bilingual Evaluation Understudy (BLEU) with (M = 0.727, SD = 0.33), Latent Semantic Analysis (LSA similarity) with (M = 0.696, SD = 0.152), and Recall Oriented Understudy for Gisting Evaluation (ROUGE) with a precision, recall and F-measure scores of (M = 0.666, SD = 0.217), (M = 0.610, SD = 0.211), (M = 0.618, SD = 0.192) for rouge-1, (M = 0.488, SD = 0.264), (M = 0.448, SD = 0.257), M = 0.452, SD = 0.248) for rouge-2 and (M = 0.602, SD = 0.241), (M = 0.553, SD = 0.235), (M = 0.5560, SD = 0.220) for rouge-L, respectively. This demonstrates its potential to expedite investigations by promptly identifying probable causes from analysis narratives, thus bolstering aviation safety protocols. Full article

Abnormal phenomena on urban roads, including uneven surfaces, garbage, traffic congestion, floods, fallen trees, fires, and traffic accidents, present significant risks to public safety and infrastructure, necessitating real-time monitoring and early warning systems. This study develops Urban Road Anomaly Visual Large Language Models (URA-VLMs), a generative AI-based framework designed for the monitoring of diverse urban road anomalies. The InternVL was selected as a foundational model due to its adaptability for this monitoring purpose. The URA-VLMs framework features dedicated modules for anomaly detection, flood depth estimation, and safety level assessment, utilizing multi-step prompting and retrieval-augmented generation (RAG) for precise and adaptive analysis. A comprehensive dataset of 3034 annotated images depicting various urban road scenarios was developed to evaluate the models. Experimental results demonstrate the system’s effectiveness, achieving an overall anomaly detection accuracy of 93.20%, outperforming state-of-the-art models such as InternVL2.5 and ResNet34. By facilitating early detection and real-time decision-making, this generative AI approach offers a scalable and robust solution that contributes to a smarter, safer road environment. Full article

This article aims to present a structured literature review that utilizes computational intelligence techniques, specifically fuzzy logic, neural networks, genetic algorithms, and machine learning, to assist in the assessment of workplace risk from human factors. The general aim is to highlight the existing literature on the subject, while the specific goal of the research is to attempt to answer research questions that emerge after the review and classification of the literature, which are aspects that have not previously been addressed. The methodology for retrieving relevant articles involved a keyword search in the Scopus database. The results from the search were filtered based on the selected criteria. The research spans a 40-year period, from 1984 to 2024. After filtering, 296 articles relevant to the topic were identified. Statistical analysis highlights fuzzy systems as the technique with the highest representation (163 articles), followed by neural networks (81 articles), with machine learning and genetic algorithms ranking next (25 and 20 articles, respectively). The main conclusions indicate that the primary sectors utilizing these techniques are industry, transportation, construction, and cross-sectoral models and techniques that are applicable to multiple occupational fields. An additional finding is the reasoning behind researchers’ preference for fuzzy systems over neural networks, primarily due to the availability or lack of accident databases. The review also highlighted gaps in the literature requiring further research. The assessment of occupational risk continues to present numerous challenges, and the future trend suggests that fuzzy systems and machine learning may be prominent. Full article

To address the issues of low efficiency in manual processing and lack of accuracy in judgment within traditional mine gas safety inspections, this paper designs and implements the Intelligent Mine Gas State Decision-Making System based on large language models (LLMs) and a multi-agent system. The system aims to enhance the accuracy of gas over-limit alarms and improve the efficiency of generating judgment reports. The system integrates the reasoning capabilities of LLMs and optimizes task allocation and execution efficiency of agents through the study of the hybrid multi-agent orchestration algorithm. Furthermore, the system establishes a comprehensive gas risk assessment knowledge base, encompassing historical alarm data, real-time monitoring data, alarm judgment criteria, treatment methods, and relevant policies and regulations. Additionally, the system incorporates several technologies, including retrieval-augmented generation based on human feedback mechanisms, tool management, prompt engineering, and asynchronous processing, which further enhance the application performance of the LLM in the gas status judgment system. Experimental results indicate that the system effectively improves the efficiency of gas alarm processing and the quality of judgment reports in coal mines, providing solid technical support for accident prevention and management in mining operations. Full article

The construction industry faces significant challenges with frequent accidents, largely due to the inefficient use of safety guidelines. These guidelines, which are often text and figure heavy, demand substantial human effort to identify the most relevant items for specific tasks and conditions. Additionally, the guidelines contain both central and peripheral elements, and central items are critical yet difficult to identify without extensive domain knowledge. This study proposes a novel recommendation framework to enhance the usability of these safety guidelines. By leveraging natural language processing (NLP) and knowledge graph (KG) modeling techniques, unstructured safety texts are transformed into a structured, interconnected KG. The PageRank and Louvain Clustering algorithm is then employed to rank guidelines by their relevance and importance. A case study on “High-rise Building Construction (General) Safety and Health Guidelines”, using ‘scaffolding’ as the keyword, demonstrates the framework’s effectiveness in improving retrieval efficiency and practical application. The analysis highlighted key clusters such as ‘fall’, ‘drop’, and ‘scaffolding’, with critical safety measures identified through their interconnections. This research not only overcomes the fragmentation of safety management documents but also contributes to advancing hazard analysis and risk prevention practices in construction management. Full article

Background: Pancreas transplantation (PTX) remains the most effective treatment to prevent long-term complications and provide consistent euglycemia in patients with endocrine pancreatic insufficiency, mainly in type I diabetic patients. Considering early graft loss (EGL) and the perioperative complication rate, an optimal risk stratification based on donor risk factors is paramount. Methods: In our single-center study, we retrospectively assessed the risk factors for EGL and reduced graft survival in 97 PTXs (82 simultaneous pancreas and kidney [SPK], 11 pancreases transplanted after kidney [PAK] and 4 pancreases transplanted alone [PTA]) between 2010 and 2021. By statistically analyzing the incorporation of different donor risk factors using the Kaplan–Meier method and a log-rank test, we introduced a composite risk model for the evaluation of offered pancreas grafts. Results: The overall EGL rate was 6.5%. In the univariate analysis of donor characteristics, age > 45 years, BMI > 25 kg/m², lipase > 60 U/L, cerebrovascular accident (CVA) as the cause of death, mechanical cardiopulmonary resuscitation (mCPR), cold ischemia time (CIT) > 600 min and retrieval by another center were identified as potential risk factors; however, they lacked statistical significance. In a multivariate model, age > 45 years (HR 2.05, p = 0.355), BMI > 25 kg/m² (HR 3.18, p = 0.051), lipase > 60 U/L (HR 2.32, p = 0.148), mCPR (HR 8.62, p < 0.0001) and CIT > 600 min (HR 1.89, p = 0.142) had the greatest impact on pancreas graft survival. We subsumed these factors in a composite risk model. The combination of three risk factors increased the rate of EGL significantly (p = 0.003). Comparing the pancreas graft survival curves for ≥3 risk factors to <3 risk factors in a Kaplan–Meier model revealed significant inferiority in the pancreas graft survival rate (p = 0.029). Conclusions: When evaluating a potential donor organ, grafts with a combination of three or more risk factors should only be accepted after careful consideration to reduce the risk of EGL and to significantly improve outcomes after PTX. Full article

Ship pollution accidents can cause serious harm to marine ecosystems and economic development. This study proposes a ship pollution accident analysis method based on a knowledge graph to solve the problem that complex accident information is challenging to present clearly. Based on the information of 411 ship pollution accidents along the coast of China, the Word2vec’s word vector models, BERT–BiLSTM–CRF model and BiLSTM–CRF model, were applied to extract entities and relations, and the Neo4j graph database was used for knowledge graph data storage and visualization. Furthermore, the case information retrieval and cause correlation of ship pollution accidents were analyzed by a knowledge graph. This method established 3928 valid entities and 5793 valid relationships, and the extraction accuracy of the entities and relationships was 79.45% and 82.47%, respectively. In addition, through visualization and Cypher language queries, we can clearly understand the logical relationship between accidents and causes and quickly retrieve relevant information. Using the centrality algorithm, we can analyze the degree of influence between accident causes and put forward targeted measures based on the relevant causes, which will help improve accident prevention and emergency response capabilities and strengthen marine environmental protection. Full article

Before starting any construction work, providing workers with awareness about past similar accident cases is effective in preventing mishaps. Based on construction accident reports, this study developed two models to identify past accidents at sites with similar site information. The site information includes 16 parameters, such as type of work, type of accident, the work in which the accident occurred, weather conditions, contract conditions, type of work, etc. The first model, the site information classification model, uses named entity recognition tasks to classify site information, which is extracted from accident reports. The second model, the similar-site accident retrieval model, which finds the most similar accidents that occurred in the past from input site information, uses a semantic textual similarity task to match the classified information with it. A total of 17,707 accident reports from South Korean construction sites were found; these models were trained to use Korean Language Understanding Evaluation–Bidirectional Encoder Representations from Transformers (KLUE-BERT) for processing. The first model achieved an average accuracy of 0.928, and the second model was precisely matched, with a mean cosine similarity score exceeding 0.90. These models could identify and provide workers with similar past accidents, enabling proactive safety measures, such as site-specific hazard identification and worker education, thereby allowing recognition of construction safety risks before starting work. By integrating site information with historical data, the models offer an effective approach to improving construction safety. Full article

Driving while drowsy poses significant risks, including reduced cognitive function and the potential for accidents, which can lead to severe consequences such as trauma, economic losses, injuries, or death. The use of artificial intelligence can enable effective detection of driver drowsiness, helping to prevent accidents and enhance driver performance. This research aims to address the crucial need for real-time and accurate drowsiness detection to mitigate the impact of fatigue-related accidents. Leveraging ultra-wideband radar data collected over five minutes, the dataset was segmented into one-minute chunks and transformed into grayscale images. Spatial features are retrieved from the images using a two-dimensional Convolutional Neural Network. Following that, these features were used to train and test multiple machine learning classifiers. The ensemble classifier RF-XGB-SVM, which combines Random Forest, XGBoost, and Support Vector Machine using a hard voting criterion, performed admirably with an accuracy of 96.6%. Additionally, the proposed approach was validated with a robust k-fold score of 97% and a standard deviation of 0.018, demonstrating significant results. The dataset is augmented using Generative Adversarial Networks, resulting in improved accuracies for all models. Among them, the RF-XGB-SVM model outperformed the rest with an accuracy score of 99.58%. Full article

Prefabricated construction is expanding and taking over traditional construction with more intervention of prefabricated building elements. Despite prefabricated construction reducing health and safety risks compared to conventional construction, there is still a risk that needs to be addressed. This article aims to investigate prefabricated construction safety through accident analysis. The accident data was retrieved through governmental resources and covered accident claims, safety costs, vulnerable occupations, and injuries (including type, cause, prior activity, and site of injury). Prefabricated construction safety is then simplistic and predictively modelled. The most common trend has been reported with graphical representation and relevant discussion. Furthermore, the trends are forecasted by using the ARIMA model (p, d, q) based on key performance parameters. The conclusion has been driven by the current status of prefabricated construction safety. This study is a pioneer in modelling prefabricated construction safety to enhance understanding of accidents and forecasting through optimization. Full article

In this study, we developed a similar text retrieval system using Sentence-BERT (SBERT) for our database of closed medical malpractice claims and investigated its retrieval accuracy. We assigned each case in the database a short Japanese summary of the accident as well as two labels: the category was classified as a hospital department mainly, and the process indicated a failed medical procedure. We evaluated the accuracies of a similar text retrieval system with the two labels using three different multilabel evaluation metrics. For the encoders of SBERT, we employed two pretrained BERT models, UTH-BERT and NICT-BERT, that were trained on huge Japanese corpora, and we performed iterative optimization to train the SBERTs. The accuracies of the similar text retrieval systems using the trained SBERTs were more than 15 points higher than those of the Okapi BM25 system and the pretrained SBERT system. Full article

Black ice has recently been identified as a major cause of transportation accidents due to detecting difficulties on the road surface. It is crucial to provide traffic users with black ice warnings beforehand to sustain commuting safety. The identification of black ice, however, is a difficult initiative, since it necessitates the installation of sophisticated monitoring stations and demands frequently manual inspection. In order to build an economical automatic black ice detection technique, the datasets are built upon a variety of weather conditions, including clear, snowy, rainy, and foggy conditions, as well as two distinct forms of pavement: asphalt and concrete pavement. The Mask R-CNN model was performed to construct the black ice detection via image segmentation. The deep learning architecture was constructed based on pre-trained convolutional neural network models (ResNetV2) for black ice detection purposes. Different pretrained models and architecture (Yolov4) were then compared to determine which is superior for image segmentation of black ice. Afterward, through the retrieved bounding box data, the degree of danger area is determined based on the number of segmentation pixels. In general, the training results confirm the feasibility of the black ice detection method via the deep learning technique. Within “Clear” weather conditions, the detecting precision can be achieved up to 92.5%. The results also show that the increase in the number of weather types leads to a noticeable reduction in the training precision. Overall, the proposed image segmentation method is capable of real-time detection and can caution commuters of black ice in advance. Full article

Cognitive failures at the information acquiring (safety training), comprehension, or application stages led to near-miss or accidents on-site. The previous studies rarely considered the cognitive processes of two different kinds of construction safety training. Cognitive processes are a series of chemical and electrical brain impulses that allow you to perceive your surroundings and acquire knowledge. Additionally, their attention was more inclined toward the worker’s behavior during hazard identification on-site while on duty. A study is proposed to fill the knowledge gap by developing the mechanism models of the two safety training approaches. The mechanism models were developed based on cognitive psychology and Bloom’s taxonomy and six steps of cognitive learning theory. A worker’s safety training is vital in acquiring, storing, retrieving, and utilizing the appropriate information for hazard identification on-site. It is assumed that those trained by advanced techniques may quickly identify and avoid hazards on construction sites because of the fundamental nature of the training, and when they come across threats, they may promptly use their working memory and prevent them, especially for more complex projects. The main benefit of making such a model, from a cognitive point of view, is that it can help us learn more about the mental processes of two different types of construction safety training, and it can also help us come up with specific management suggestions to make up for the approaches’ flaws. Future research will concentrate on the organizational aspects and other cognitive failures that could lead to accidents. Full article