CNN-driven self-powered sensing system for transmission line aeolian vibration recognition by hybrid aeolian energy harvester

Actor-critic guided CDBN with GAN augmentation for robust facial emotion recognition.

Dynamic facial expressions of emotion are typically recognized more accurately than static expressions, especially among individuals with immature, vulnerable or impaired facial expression recognition (FER) systems, such as children, older adults, and individuals with clinical conditions. These findings underscore the need of assessing both dynamic and static stimulus formats and suggest that the dynamic advantage could serve as a potential individual-level marker of FER impairment. However, previous research has primarily focused on group-level effects, often overlooking critical individual differences. Here, we tested whether the QUEST threshold-seeking algorithm can efficiently estimate the minimal signal required for accurate recognition of static and dynamic facial expressions at the individual level. We also quantified the minimum number of trials needed to obtain stable threshold estimates. To evaluate its sensitivity, we compared FER thresholds across neurotypical young adults, older adults, and a well-documented case of acquired prosopagnosia. Our findings demonstrate that the QUEST algorithm is a robust and efficient tool for rapidly estimating meaningful FER thresholds at the individual level. We also provide guidance on the minimum number of trials required to obtain stable threshold estimates and identify the facial expressions that serve as the most sensitive markers for probing the dynamic advantage. This psychophysical approach is particularly well-suited for single-subject analyses, assessment of FER in populations with limited capacities, and inclusion in comprehensive testing batteries. Collectively, this psychophysical approach enables scalable, time-efficient screening and longitudinal monitoring of FER, facilitating cross-individual and population comparisons in both research and clinical settings.

Medical Entity Recognition (MedER) systems are needed to enhance the use and accessibility of Natural Language Processing (NLP) methods in the medical field. Since medical entity recognition in Bangla is a relatively new field, no such datasets are currently available in any repository. Unlike AI-generated data, which may contain biases or errors from auto- mated algorithms, the Bangla-MedER dataset is a manually curated resource for multi-type medical entity recognition in Bangla-language drug-indication text. A total of 2980 records were collected from publicly available medicine-related websites, pharmaceutical articles, and other sources of drug information. Each record contains the original Bangla medical text, along with expert-verified annotations across six entity types: medicine/chemical name, organ, disease, hormone, pharmacological class, and common medical terms. The raw transcribed text is provided to support reproducible research. All annotations were performed manually under the guidance of a certified medical expert, and proprietary brand names or personally identifiable information were removed to ensure privacy. The Bangla-MedER dataset enables a variety of applications, including medical named entity recognition in Bengali medical text, biomedical information retrieval, and clinical decision-support systems in a low-resource language environment. The complete raw dataset, along with documentation, is publicly available, offering a benchmark for medical entity recognition, healthcare NLP, biomedical informatics, and another related research.

Association between EMS response time and return of spontaneous circulation in out-of-hospital cardiac arrest patients in Busan, South Korea.

Towards a lightweight YOLOv8n for aquaculture feeding detection: Architectural improvements for feature enhancement and computational efficiency

Colonoscopy report generation using voice recognition system.

Dance motion recognition and correction present unique challenges due to the subtle distinctions in movement execution that differentiate correct performance from flawed attempts. This paper proposes an integrated system combining 3D motion capture technology with a novel spatiotemporal attention-based graph convolutional network for accurate dance action recognition and intelligent feedback generation. The proposed architecture features a dual-stream design incorporating adaptive graph topology learning that discovers task-relevant relationships between non-adjacent joints, alongside multi-scale temporal modeling to capture movement dynamics across varying time scales. A multi-dimensional correction feedback algorithm translates recognition outputs into prioritized, actionable guidance by comparing performer movements against professional reference templates through dynamic time warping alignment and joint-level deviation analysis. We constructed DanceMotion-86, a comprehensive dataset comprising 10,836 clips across 86 action categories spanning five dance genres. Experimental results demonstrate that the proposed method achieves 92.3% recognition accuracy, outperforming state-of-the-art baseline methods. User studies with 36 participants confirmed 87.4% error detection rate and showed significantly accelerated skill acquisition among feedback-enabled learners compared to control conditions. The system offers practical applications for intelligent dance instruction, cultural heritage preservation, and remote training platforms.

The ever-expanding biomedical literature necessitates an efficient and robust mining platform, with the foundational step being a reliable Biomedical Named Entity Recognition (BioNER) system. Existing approaches, such as multi-task and collaborative learning, have attempted to address dataset heterogeneity but often rely on complex architectures with task-specific layers, limiting scalability. A key research gap is the development of a unified model that optimises across multiple datasets without sacrificing performance or introducing architectural complexity. In this study, we propose a novel Loss-Masking Optimisation framework for BioNER models that enables multi-dataset training via a dataset-aware masking strategy. This approach extends the standard BERT-based NER pipeline by introducing a tag-masking array that nullifies logits for tags absent in the originating dataset, thereby reducing cross-dataset interference. Using this methodology, we trained a single BioNER model across all 16 biomedical NER datasets, achieving higher precision and overall F1 scores than conventional multi-dataset training. While some datasets showed performance gains, others stayed near baseline, and a few declined, underscoring the nuanced impact of dataset interactions. To the best of our knowledge, this is among the first studies to apply a dataset-aware loss-masking mechanism to unified multi-dataset BioNER training, offering a scalable alternative to multi-task architectures.

Emotion recognition is an important area in artificial intelligence and human-computer interaction. Traditional emotion detection systems mainly depend on a single source such as facial expressions or speech signals, which may reduce accuracy in real-time conditions. This paper proposes a multimodal emotion detection framework using audio, video, and text-based expressions with artificial intelligence techniques. The proposed system uses Convolutional Neural Networks (CNN) for facial emotion recognition, Long Short-Term Memory (LSTM) networks for speech emotion analysis, and Natural Language Processing (NLP) for understanding text queries. The system processes multimedia inputs and predicts emotions such as happiness, sadness, anger, fear, surprise, and neutral emotions. The framework is implemented using Python, TensorFlow, Keras, OpenCV, and Flask. Experimental results show that combining multiple modalities improves emotion recognition accuracy and system performance compared to traditional single-modality methods.

Metal oxide semiconductor (MOS) gas sensors hold great promise for gas detection due to their low cost and miniaturization. Nevertheless, such sensors are unable to distinguish structurally analogous gas molecules due to mere reliance on one-dimensional resistance signals. The construction of a sensor array offers a feasible strategy to improve gas-sensing selectivity by increasing signal dimensionality, while it inevitably increases system complexity and cost, limiting their practical applicability. In this work, we propose a dynamic light-pulse modulation strategy that can expand the signal dimensions based on a single chemiresistive gas sensor, achieving the selective identification of structurally similar volatile organic compounds (VOCs). Using Ag-modified ZnO nanocomposites as the sensing materials, the fabricated gas sensor exhibits favorable response and recovery characteristics toward formic acid under optimal working conditions, yet it also presents highly analogous response behaviors toward structurally similar ethanol and acetic acid. Employing periodic light pulses to the sensor, multidimensional feature parameters are extracted from the transient resistance in different gas environments. By further integrating both dynamic and steady-state parameters into a multidimensional feature set and applying a support vector machine classifier, highly selective recognition of structurally similar VOCs (including formic acid, ethanol, and acetic acid) can be achieved across a wide concentration range, with an identification accuracy rate reaching as high as 97.8%. The proposed strategy combining dynamic light-pulse modulation and machine learning provides a valuable pathway to overcome the cross-sensitivity of MOS sensors, laying a foundation for the development of compact, intelligent gas recognition systems for practical applications.

Traffic Sign Recognition System with Voice Alerts Using IoT Technology

Integrating stereoselective synthesis and molecular-behavioral approaches to characterize aggregation pheromone in the German cockroach.

Facial emotion recognition (FER) plays a vital role in understanding human behavior and communications, with applications in human-computer interaction, surveillance, healthcare, and multimedia content analysis. Emotion recognition is challenging as it necessitates capturing and analysing subtle variations in facial expressions together with dynamic speech variations. Hearing impairment in youngsters and children has significant effects on social, emotional, and behavioural growth. Communication problems could affect the social and emotional improvement of a hearing-impaired individual. Emotion is a very regular incident of all human beings, whether impaired or normal. Usually, emotions are identified by facial expressions. FER is the most influential, nonverbal, and natural means for individuals to convey their strength and emotions. Automatic FER has received great attention currently. FER is a significant area in the domains of artificial intelligence (AI) and computer vision (CV) because of its important commercial and academic potential. The facial expression detection is challenging for machine learning (ML) methods; meanwhile, people can differ considerably in the manner in which they show their expression. This article develops a Context-Aware Interaction with Fusion Feature Models for Enhanced Facial Emotion Recognition (CAIFFM-EFER) approach in disabled hearing individuals. The main purpose of this article is to develop and calculate a model for accurate and real-time FER by applying advanced techniques. At first, the image processing stage employs the Wiener filter (WF) to enhance image quality by eliminating the noise. For the feature representation process, the CAIFFM-EFER method employs a fusion of VGG-19, MobileNetV1, and InceptionNetV3 models. Followed by, the stacked sparse autoencoder (SSAE) approach is used for facial emotion classification. Finally, context-aware interactions driven by large language models (LLMs) facilitate more refined reasoning and generate adaptive, context-sensitive responses. The experimentation evaluation of the CAIFFM-EFER model portrayed a superior accuracy value of 99.27% over existing methods under the Emotion detection dataset.

The proposed Smart Surveillance System presents a novel, hardware-integrated prototype demonstration aimed The proposed Smart Surveillance System presents a groundbreaking hardware-integrated prototype that decisively validates the effectiveness of a dual-branch anti-spoofing model on the low-power edge device, Raspberry Pi 3B+. This prototype goes beyond algorithmic performance research, showcasing a fully functional proof of concept. In contrast to existing surveillance solutions that typically rely on centralized cloud processing or basic recognition systems, our system employs an advanced dual-branch model that utilizes both spatial and frequency-domain features. This approach enables real-time anti-spoofing with an impressive error rate of less than 2%. What truly sets our system apart is its seamless end-to-end integration of cloud-based authentication, edge-level inference for rapid response, and an interactive live video conferencing feature. This configuration empowers immediate verification and action during potential spoofing events. With IoT-enabled devices, our system ensures effortless communication for live streaming, automated alerts, and scalable cloud data management. Coupled with edge computing, it guarantees real-time decision-making with minimal latency. Experimental results confirm its high accuracy in distinguishing genuine users from spoofing attempts, positioning our solution as a lightweight, proactive, and user-interactive surveillance option that is perfectly suited for homes, enterprises, and public infrastructures.

This paper presents an AI-driven multisensor wearable system for real-time breathing pattern recognition by integrating an inertial measurement unit (IMU) and a flex sensor with wireless data connectivity. Three artificial intelligence models-transformer, convolutional neural network-long short-term memory (CNN-LSTM), and histogram gradient boosting (HGB)-were evaluated for breathing pattern recognition across different model complexities (complex, simple, pure) and sensor configurations (IMU , Flex , and combined). The multisensor system combined with AI model was tested with multiple participants. The complex transformer model, trained with focal loss on the combined IMU and flex sensor data, achieved the highest performance, with 93.41% accuracy and a mean area under the curve (AUC) of 0.9919, outperforming all the other models. Multimodal input significantly improved classification accuracy-up to 20% higher than flex sensor models in six-class tasks-while focal loss enhanced robustness, particularly in addressing class imbalance. These results demonstrate the potential of combining wearable sensor fusion with deep learning to enable accurate, noninvasive, and wireless real-time respiratory monitoring, with potential applications in clinical diagnostics, telemedicine, and personalized health tracking.

Emotions play a vital role in human communication and contribute significantly to building a healthy and socially connected society. In recent years, researchers have developed various Facial Expression Recognition (FER) systems to interpret human emotions as the demand for advanced human–computer interaction applications continues to grow. However, creating an FER system that performs effectively in real-world environments remains a challenging task due to variations in lighting, pose, and other uncontrolled imaging conditions. This research work presents an FER system that utilizes both machine learning and deep learning techniques to recognize emotions from static images as well as video sequences. Within the machine learning framework, two novel methods are introduced: the Statistical Shape Projection Model (SSPM) and the Enhanced Multi-Feature Fusion Model (EMFM), which focus on geometric and appearance-based feature extraction methods respectively. In the deep learning framework, a Maximum Boosted Convolutional Neural Network (MBCNN) and a hybrid architecture combining MBCNN with Long Short-Term Memory (LSTM) are proposed to improve the accuracy of emotion classification. The Statistical Shape Projection Model (SSPM) achieves high classification performance by extracting meaningful features using a statistical shape model combined with integral projection analysis. In this approach, a static image is taken as input and its edge information is enhanced through the use of un sharp masking. The statistical shape model adapts the new input image to the trained model structure, allowing the deformation patterns to be learned efficiently. The effectiveness of the proposed feature extraction method is further improved when combined with projection-based analysis. Additionally, an edge enhancement technique highlights prominent structural details, enabling the system to identify discriminative features more effectively for accurate facial expression recognition

Fingerprint-based authentication systems (FAS) play a crucial role in secure access control, including academic libraries. Conventional fingerprint recognition systems that rely on a single feature extraction technique often struggle to extract robust features, leading to high false positive rates and low accuracy. This research developed a feature-fusion authentication system for academic library access control using multi-feature extraction techniques. 324 university students fingerprint dataset from 81 subjects were captured. The acquired dataset was preprocessed (cropped, contrast adjustment, gray scale, binarization). The Cross Number Algorithm (CNA) and Principal Component Analysis (PCA) were used for feature extraction. The Weighted Sum Rule was used to fuse extracted features from CNA and PCA, generating a unified feature vector. Random Forest Classifier was employed for classification. The results show that CNA–PCA based system achieved accuracy of 96.91%, CNA achieved accuracy of 94.14% and PCA produced accuracy (92.59%).

Specificity, the selective partnership between a host and particular microbial taxa, is a fundamental feature of microbial symbioses, yet the mechanisms that generate and maintain specificity can be difficult to disentangle across the evolutionary, ecological and molecular scales at which they operate. Binary symbioses, in which a single host and microbial species interact, offer powerful systems for investigating these mechanisms across biological scales. Here, current knowledge of the symbiosis between siphonfish in the genus Siphamia and their bioluminescent symbiont, Photobacterium mandapamensis, is synthesized to illustrate ways in which specificity operates across multiple scales in this vertebrate-bacteria association. At the evolutionary scale, P. mandapamensis is the exclusive symbiont across all Siphamia species throughout the Indo-Pacific examined to date, indicating specificity is a conserved feature of the association. At the ecological scale, host behavior may generate local symbiont pools that reinforce specificity across host generations, promoting fine-scale genetic divergence among symbiont populations. At the molecular scale, comparative genomics between P. mandapamensis and the closely related, yet incompatible P. leiognathi reveals candidate loci unique to P. mandapamensis that encode putative systems for exopolysaccharide biosynthesis, iron transport, host-specific attachment and surface recognition, and nitrogen assimilation. Together, these findings illustrate that specificity in this system is not the product of any single mechanism, but of multiple processes operating across these scales and feeding back to one another, positioning the Siphamia-P. mandapamensis symbiosis as a tractable model for investigating how partner fidelity is generated, maintained, and potentially disrupted in a changing world.

Robot therapy, with PARO being a prominent, has become an effective intervention in elderly care, where caregivers play an important role in patient engagement during therapy. This study analyzes caregiver behavior to reveal behavioral parameters associated with caregiving skill in delivering robot therapy using PARO. The analysis parameters included physical activity of caregivers and patients during therapy, facial gaze, eye level between caregiver and patient, and PARO face orientation. Using a vision-based activity recognition system, observational data were collected from 24 novice caregivers (12 trained, 12 untrained) and 6 elderly participants, with statistical analysis (Mann-Whitney U tests) confirming significant differences in interaction behavior between trained and untrained caregivers. The results showed that trained caregivers demonstrated specific technical skills, including holding PARO longer during introductions, delaying the initial placement of PARO, sustaining facial gaze, maintaining longer eye contact, and dynamically adjusting PARO's position based on patient cues. Although trained caregivers tended to maintain a smaller eye-level difference with patients on average, this difference was not statistically significant. These findings indicate that formal caregiving education is associated with systematic differences in interaction behavior, even at a pre-professional stage. The study contributes a quantitative framework for identifying foundational behavioral markers related to caregiving skill in robot therapy, rather than evaluating clinical efficacy.