Browse Articles
Discover research articles across all indexed journals
Optimal features assisted multi-attention fusion for robust fire recognition in adverse conditions
Abstract Deep neural networks have significantly enhanced visual data-based fire detection systems. However, high false alarm rates, shallow-layered networks, and poor recognition in challenging environments continue to hinder their practical deployment. To address these limitations, we introduce the Attention-Enhanced Fire Recognition Network (AEFRN). This novel progressive attention-over-attention framework achieves state-of-the-art (SOTA) performance while maintaining computational efficiency. Our approach introduces three key innovations: Firstly, Convolutional Self-Attention (CSA), integrating global self-attention with convolution through dynamic kernels and trainable filters for enhanced low-level fire feature processing. Secondly, Recursive Atrous Self-Attention (RASA) with optimized dilation rates, capturing comprehensive multi-scale contextual information through a recursive formulation with minimal parameter overhead. Thirdly, an enhanced Convolutional Block Attention Module (CBAM) with modified channel and spatial attention mechanisms for robust feature discrimination. We validate AEFRN’s interpretability using Grad-CAM visualization, demonstrating effective attention focus on fire-relevant regions. Comprehensive experimental evaluation on FD and BoWFire benchmark datasets shows AEFRN’s superiority over SOTA methods, achieving 99.11% accuracy on the FD dataset, and 97.98% accuracy on the BoWFire dataset. Extensive comparisons against twelve SOTA approaches confirm AEFRN’s effectiveness for fire detection in challenging scenarios while maintaining computational efficiency suitable for practical deployment.
Integrating AI predictive analytics with naturopathic and yoga-based interventions in a data-driven preventive model to improve maternal mental health and pregnancy outcomes
Abstract Maternal mental health during pregnancy is a crucial area of research due to its profound impact on both maternal and child well-being. This paper proposes a comprehensive approach to predicting and monitoring psychological health risks in pregnant women using advanced machine learning techniques. The study employs a systematic methodology including data collection, preprocessing, feature selection, and model implementation. Data collection was conducted at Majidia Hospital, involving a diverse sample of 70,000 pregnant women recruited through antenatal clinics, online health platforms, community outreach programs, and telephone surveys using structured questionnaires. Participants were selected across all pregnancy trimesters to ensure a representative demographic, capturing variations in age, educational background, occupational status, and parity. A diverse set of machine learning models, including Random Forest, Decision Tree, Support Vector Machine (SVM), Logistic Regression, Gaussian Naive Bayes, and Multilayer Perceptron (MLP), were evaluated alongside ensemble methods to achieve robust and reliable predictions. The experimental results demonstrate that the Random Forest model consistently outperforms other classifiers with an accuracy of 97.82% ± 0.03%, precision of 97.82% ± 0.03%, recall of 100.00% ± 0.00%, and an F1 score of 96.81% ± 0.02%. SVM and Decision Tree classifiers also showed strong performance, with accuracy scores of 93.79% ± 0.01% and 91.82% ± 0.03%, respectively. Furthermore, ensemble methods enhanced predictive performance, highlighting their ability to balance accuracy, precision, recall, and F1 score. In regression tasks, the Random Forest Regressor achieved near-perfect predictions with a Mean Squared Error (MSE) of 4.5767 × 10−8 and an R2 score of 1.000, underscoring its superior predictive capabilities. Additionally, a custom loss function integrating Cross-Entropy Loss and an F1 Score Penalty was introduced to address class imbalance and enhance model performance. The training process, conducted over 10 epochs, demonstrated consistent loss reduction, with the lowest recorded loss at epoch 8 (2.4382), reflecting effective learning and parameter tuning. This study envisions the development of an intelligent, web-based tool aimed at revolutionizing psychological health assessment and support for pregnant women. This tool will not only provide early diagnosis and intervention but also recommend personalized yoga practices and natural remedies to improve maternal mental health and overall wellbeing. These findings highlight the potential of AI-driven innovations in enhancing maternal care through holistic and accessible technological solutions.
Shape sensing robotic assisted bronchoscopy versus virtual bronchoscopic navigation in the diagnosis of peripheral pulmonary nodules
Machine learning analysis of a Fano resonance based plasmonic refractive index sensor using U shaped resonators
Catechin suppresses HNSC via STXBP1 dependent inhibition of macrophage infiltration and CD47 mediated immune evasion
Intelligent data-driven system for mold manufacturing using reinforcement learning and knowledge graph personalized optimization for customized production
Multi-omics and experimental validation reveal mechanism of compound mylabris capsules in treating diffuse large B-cell lymphoma
Rock blasting evaluation - image recognition method based on deep learning
Impact of acute stress exposure on genome-wide DNA methylation
Intelligent brain tumor detection using hybrid finetuned deep transfer features and ensemble machine learning algorithms
Prediction of suicide using web based voice recordings analyzed by artificial intelligence
The impact of circular economy initiatives on urban air quality in the United States
Enhancement of systemic acquired resistance in rice by F-box protein D3-mediated strigolactone/karrikin signaling
Control of biofilm from single and multispecies bacteria associated with food spoilage using metabolite of Streptomyces sp. KP110 and Pseudomonas fluorescens JB 3B
MultiFG: integrating molecular fingerprints and graph embeddings via attention mechanisms for robust drug side effect prediction
Upregulation of VEGFA through the adenosine A2A receptor is a crucial pathway for inhibiting pericyte apoptosis in chronic cerebral hypoperfusion
Abstract Chronic cerebral hypoperfusion (CCH) is a key factor in vascular cognitive impairment. Pericyte loss and subsequent blood-brain barrier disruption play pivotal roles in the pathogenesis of CCH-induced white matter lesions (CCH-WMLs). Previous work suggested that the adenosine A2A receptor (A2AR) may protect pericytes in CCH-WMLs, but the mechanisms are not fully understood. In this study, we induced CCH in Sprague‒Dawley rats via bilateral carotid artery occlusion and treated them with the A2AR agonist CGS21680 or the A2AR antagonist SCH58261. Our findings revealed that CGS21680 significantly inhibited the expression of the proapoptotic proteins BAX and Caspase 3, while SCH58261 obviously promoted it. The expression of the antiapoptotic protein Bcl-2 was markedly increased by CGS21680 in OGD-exposed pericytes. Additionally, the expression of the transcription factors Rap-1, ERK, and phosphorylated ERK also increased dramatically in OGD-exposed pericytes following CGS21680 administration. VEGFA and VEGFR2 expression was upregulated by CGS21680 and downregulated by SCH58261 in pericytes after OGD. Furthermore, VEGFA knockdown via a shRNA-expressing adenovirus counteracted the protective effect of A2AR against pericyte apoptosis following OGD. Notably, the expression of BAX and Caspase3 was significantly upregulated, and the expression of BCL-2 was markedly downregulated in OGD-exposed pericytes after Rap-1 knockdown via a shRNA-expressing adenovirus. Rap-1 suppression obviously reduced the levels of phosphorylated ERK, VEGFA and VEGFR2 in pericytes, suggesting a role for the Rap1-ERK pathway in the A2AR-induced upregulation of VEGFA expression. Overall, A2AR activation inhibits pericyte apoptosis and may exert neuroprotective effects against CCH by increasing VEGFA expression through the Rap1-ERK signaling pathway.
Photonic crystal biosensor featuring an eye-shaped cavity for precise identification of cancerous cells
Abstract This article presents a highly sensitive and thermally stable photonic crystal (PhC) biosensor designed for accurate cancer-cell detection. The proposed sensor features a square-lattice of silicon rods (radius 0.1 µm) with a photonic bandgap spanning 1.2–2.1 µm. It includes two line-defect waveguides for input and output, and a uniquely engineered Eye-shaped cavity that holds the analyte as embedded rods. These rods are strategically arranged along the Eye-shaped boundary and the central area resembling an iris, facilitating accurate detection through resonance wavelength shifts triggered by changes in the refractive index. The biosensor demonstrates excellent transmission efficiency (69.7%–99.9%), high sensitivity (236–243 nm/RIU), and a strong quality factor (15,764–87,070), ensuring sharp and clearly defined resonance peaks. A key advantage of the design is its linear response to refractive index variations, which enhances detection accuracy and supports reliable real-time biosensing. Moreover, the sensor maintains stable performance across a wide temperature range (25 °C to 75 °C) and exhibits robust tolerance to fabrication variations. These features validate the biosensor’s precision for biomedical diagnostics.
Role of PHB2 as a potential biomarker in pan-cancer: a multi-database analysis
Metal loss defect detection and depth estimation using multi-spectral image analysis of cooling excited steel specimen with corrosion
Abstract Imaging techniques have considerably improved corrosion-induced metal loss defect detection and severity estimation in recent decades. Even though the detection of defects using imaging techniques in steel is well established, determining the severity remains difficult due to the necessity of estimating the depth information of the defect from 2-dimensional image data. This study used a steel test specimen with artificial defects of varying depths and diameters, subjected to accelerated corrosion. A Multi-Spectral Imaging setup observed the specimen’s spectral response at different temperatures following a cooling excitation. Reflected intensities at specific wavelengths indicated defect presence and allowed quantification of corrosion-induced metal loss. Principal Component Analysis and machine learning regression were used to transform discrete defect depths into continuous assessments. Support Vector Regression, Decision Tree Regressor, Random Forest Regressor, Gradient Boosting Regressor, and a Feedforward Neural Network (FNN) were tested for this task. The FNN showed the best results in solving the regression problem with a least Root Mean Square Error of 0.2829 and an R2 score 0.976. The 700 nm–900 nm range was identified as the optimal wavelength span for spectral imaging.