Photoplethysmography (PPG) is commonly used to gather health-related information but is highly affected by motion artifacts from daily activities. Inspired by the strong denoising capabilities and generalization of diffusion probabilistic models, this paper proposes a novel PPG denoising method using a diffusion probabilistic model to reduce the impact of these artifacts. While typical diffusion models handle Gaussian noises, motion artifacts often involve non-Gaussian noise. To address this, the proposed method incorporates noisy PPG signals into both the diffusion and reverse processes, allowing the model to adapt better to complex and non-Gaussian noises. A dataset with clean and noisy PPG signals from 15 subjects performing various motion tasks was collected for evaluation. The results show the proposed model significantly improves PPG signal quality, reducing the Peak-Rejection-Rate (PRR) from 0.24 to 0.03. It also enhances the accuracy of heart rate (HR) estimation and various heart rate variability (HRV) measures, showing robustness and good generalization across different tasks and subjects.
Existing studies have revealed that the reading comprehension ability of readers can be adversely affected by their psychosocial stress. Yet, the detailed impact of stress on various stages of text processing is understudied. This study aims to explore how the higher-level text processing ability, including syntactic parsing, sentence integration, and global text processing, of first language (L1) and second language (L2) English readers is affected under stress situations. In addition, the roles of trait anxiety, the central executive function moderating stress effects, in text processing were also examined. Twenty-two L1 readers and twenty-one L2 readers were asked to perform reading comprehension tasks under different stress situations. Eye-tracking technology was adopted to record participants’ visual behaviors while reading, and ten eye-movement measurements were computed to represent the effect of different types of text processing. The results demonstrate that the stress reduced the efficiency of syntactic parsing and sentence integration in both L1 and L2 groups, but only impaired global text processing in L2 readers. Specifically, L2 readers focused more on the topic structure of text to facilitate comprehension under stress situations. Moreover, only L1 readers’ higher-level text processing was affected by trait anxiety, while L2 readers’ processing was mainly related to their reading proficiency level. Future studies and applications were discussed. The findings advance our understanding of stress effects on different stages of higher-level text processing. They also have practical implications for developing interventions to help language learners suffering from stress disorders.
In safety-critical environments characterized by volatility, uncertainty, and complexity, operators are persistently exposed to elevated stress levels, posing significant threats to both operational safety and operator well-being. Existing AI-driven stress detection approaches rely on task-specific supervised models with limited scalability and lack the explainability and reliability required for high-stakes applications. This paper introduces a trustworthy hybrid human–agentic AI framework integrating In-Context Learning (ICL) with a Knowledge Graph-enhanced Retrieval-Augmented Generation (GraphRAG) module for explainable stress detection and personalized intervention. The ICL component enables few-shot, context-aware stress detection by fusing physiological signals with multi-contextual information, and the GraphRAG module grounds explanations and intervention recommendations in structured domain knowledge, reducing hallucination in large language models (LLM). The framework is validated through a Vessel Traffic Service (VTS) case study, using multimodal data collected from real-world operations over a five-month on-site deployment. Without task-specific training, the ICL-based predictor achieves an F1-score of 88.37% with only 256 examples, surpassing traditional machine learning baselines. GraphRAG constructs a domain-specific knowledge graph linking physiological indicators, operational contexts, stress states, and intervention strategies, enabling multi-hop reasoning for evidence-grounded, traceable recommendations. Expert evaluations confirm that GraphRAG-enhanced outputs significantly improve accuracy, usability, and trustworthiness over LLM-only baselines. This work advances occupational stress management by transitioning from passive monitoring toward a trustworthy, human-centred decision-support paradigm with strong potential for real-world deployment in safety-critical environments.