Accurate and real-time facial recognition remains a cornerstone of modern computer vision applications. However, existing systems often suffer from high computational costs, limited scalability, and poor adaptability to large, heterogeneous datasets. This paper presents a lightweight yet robust deep learning-based face recognition framework that leverages embedding-based classification using a custom-designed Deep Neural Network (DNN) architecture. The proposed system integrates Dlib’s ResNet-based face embedding extractor with a bespoke DNN classifier, trained and evaluated on both a custom 68-label dataset and the publicly available Labeled Faces dataset comprising 5,817 identities. The framework encompasses a comprehensive pipeline including face detection using HOG and Haarcascade algorithms, landmark-based alignment using 68 facial points, grayscale preprocessing, and extensive data augmentation to enhance generalization. A 128-dimensional facial encoding is used as input to the DNN, which employs dropout regularization and ReLU activation across fully connected layers to optimize classification performance. Extensive experimentation demonstrates the superiority of the proposed model over traditional and pretrained models. On the 68-label dataset, the DNN achieved 99.37% accuracy, 98% precision, and 89% F1-score, outperforming both Dlib+SVC and conventional methods such as LBPH, Eigenface, and Fisherface. Furthermore, for the large-scale 5,817-label dataset, it attained a 94% accuracy, significantly higher than the 54% achieved by the pretrained SVC model. Real-time testing using a live camera further confirms the framework’s practicality, delivering high-confidence recognition with low latency. This research not only bridges the gap between lightweight deployment and high accuracy but also paves the way for scalable and efficient face recognition in resource-constrained environments.

Maintaining optimal control of heating boiler systems using intelligent systems poses a significant challenge due to inherent nonlinearities, time delays, and unpredictable variations in fuel quality and thermal load. Conventional fuzzy logic controllers, although effective under steady-state conditions, often fail to deliver robust performance when subjected to abrupt parameter fluctuations. To address this limitation, this study proposes a novel hybrid quantum-fuzzy inference framework that enhances the adaptability and robustness of intelligent control systems for heating boiler automation. The approach integrates core quantum computing principles—specifically, superposition and probabilistic amplitude processing—with fuzzy rule-based logic to dynamically activate and generalize across a predefined set of control rules. A self-organizing and adaptive knowledge base is developed in real time, allowing the control logic to restructure itself in response to disturbances such as ±25% fluctuations in fuel flow rate, up to 30% variations in thermal demand, and 5–8 second measurement delays. The proposed control model is developed and validated using MATLAB/Simulink on a nonlinear heating boiler plant subjected to realistic operational disturbances, including random fuel composition. Simulation results indicate that the hybrid system achieves up to 36% improvement in control stability, 30% faster response time, and 22% reduction in energy consumption compared to conventional fuzzy systems. These findings demonstrate that the quantum-fuzzy approach is a promising solution for robust and energy-efficient automation of nonlinear thermal energy conversion systems, and can be generalized to other complex industrial processes requiring adaptive intelligent control.

Background: Individuals with speech impairments often face significant challenges in daily communication, limiting their ability to interact effectively. Traditional communication aids, though helpful, can be costly or inflexible. Recent advancements in computer vision and deep learning offer new opportunities to develop logical, real-time, and affordable assistive technologies. Objective: This study aims to design and implement a low-cost, vision-based gesture-to-speech system that enables nonverbal individuals to communicate through hand gestures. The goal is to translate recognized gestures into audible speech, bridging the communication gap and enhancing quality of life. Methods: The system uses a standard webcam to capture hand gestures, processed in real-time using OpenCV. A Convolutional Neural Network (CNN) developed with TensorFlow is trained on a custom dataset to classify hand signs accurately. The workflow includes image preprocessing, data augmentation, model training, and deployment. Each recognized gesture is mapped to a corresponding text, which is then converted into speech using a text-to-speech (TTS) engine. Results: The captured hand image is first passed through a filter, and the filtered image is then input to a CNN-based classifier that predicts the gesture class. Once classified, the corresponding word is displayed as output and then converted into audible speech. The system achieved 98% accuracy across 26 alphabetic gestures and performed reliably in real-time with minimal latency under varying lighting and background conditions. Conclusion: The proposed system is an effective and affordable communication aid for individuals with speech impairments. Its modular, real-time design makes it suitable for deployment in resource-constrained settings.

AI-Powered Smart Urban Navigation and Safety Alert System for Visually Impaired Pedestrians

Navigating urban environments remains a significant challenge for visually impaired individuals due to dynamic obstacles, unclear signage, and inconsistent auditory cues. This research presents an AI-powered smart urban assistive system designed to provide real-time navigation guidance and safety alerts to visually impaired pedestrians. The system integrates computer vision, sensor fusion, GPS-based routing, and audio feedback mechanisms to create a wearable and responsive mobility aid.

A prototype was deployed in both controlled campus and real-world urban scenarios involving 20 visually impaired users across various navigation tasks (e.g., crossing roads, avoiding obstacles, locating entrances). The proposed system achieved an accuracy of 93.2% in obstacle detection, 87.6% route adherence, and 95% user satisfaction in safety perception, significantly outperforming traditional cane-based mobility methods.

In conclusion, the AI-driven smart assistive system demonstrates high potential in enhancing urban mobility for the visually impaired. It empowers users with greater independence, real-time situational awareness, and reduced anxiety in navigating complex environments. The modular and scalable design ensures adaptability to various urban infrastructures and user preferences. Future work will explore integration with edge AI for offline inference, voice-command interfaces, and context-aware navigation recommendations based on pedestrian density and time of day, contributing toward inclusive smart cities in alignment with Sustainable Development Goals (SDG 11: Sustainable Cities and Communities).

Developing green materials is essential for the transition toward more sustainable engineering solutions. Among these, biopolymers and their composites represent promising alternatives to petroleum-based polymers due to their biodegradability, lower toxicity, and potential cost-efficiency—especially when derived from waste. In this work, we explore the feasibility of using a bio-based resin, epoxidized resveratrol (RESEP), reinforced with continuous basalt fibers as a sustainable composite for non-structural components in railway vehicles.

Three composite laminates were fabricated via the hand lay-up method, a reference laminate (RESEP + basalt), a laminate with 5%wt DOPO (a flame-retardant agent), and a laminate with 7.5%wt mechanically recycled carbon fiber (RCF), to obtain smart materials by developing electrically conductive composites. The composites were characterized by density measurements, mechanical and thermomechanical testing, and fire resistance evaluations. Additionally, the RCF-reinforced laminate was assessed for structural health monitoring (SHM) capabilities and de-icing performance via Joule heating.

Our results show that the RCF-reinforced laminate combines good mechanical behavior with added functionalities such as SHM and active de-icing, despite slight mechanical trade-offs. In contrast, DOPO inclusion negatively impacted mechanical and fire performance. Here, a significant decrease in interlaminar shear strength was observed, likely due to the increased resin viscosity, hindering fiber impregnation. These defects compromise both mechanical integrity and fire resistance. Consequently, the RCF laminate is proposed for external applications requiring multifunctionality, while the reference laminate is more suitable for fire-sensitive interior uses.

This study supports the advancement of sustainable, high-performance composites for the transportation sector in alignment with circular economy and emission reduction goals.

Introduction: The human mouth is a challenging environment due to the presence of bacteria, its relatively high temperature, and high moisture levels. This poses a challenge to designing materials possessing satisfactory functional properties. One of those aspects is to achieve low water sorption and water leachability, because water, which is present in humans’ saliva, can cause excessive swelling of the material and leaching out of uncured monomer.
The aim of the study is the modification of dental copolymer with urethane-dimethacrylate monomers possessing quaternary ammonium groups and the determination of their water sorption, water solubility and water contact angle.

Methods: The monomers were synthesised in a three-step process involving transesterification of MMA, then N-alkylation with alkyl bromide and finally addition to diisocyanate (1,3-bis(2-isocyanatopropan-2-yl)benzene, isophorone diisocyanate, 4,4’-methylenedicyclohexyl diisocyanate, 4,4’-diphenylmethane diisocyanate). The monomers were mixed with commercial monomers bisphenol A glycerolate dimethacrylate and triethylene glycol dimethacrylate, and the initiating system camphorquinone and N, N-dimethylaminoethyl methacrylate. The copolymers were obtained in the photopolymerization process. The reference copolymer was also prepared. The water sorption, water solubility and water contact angle were determined for copolymers.

Results: The eight modified copolymers were obtained. The modified copolymers showed higher water sorption and water leachability. Regarding the water contact angle, five copolymers showed lower values, two copolymers showed similar values, and one copolymer showed a higher value of water contact angle compared to the reference copolymer.

Conclusions: The factors that influenced copolymers' properties were the length of the N-alkyl substituent and the diisocyanate core.

Colorectal polyps are unusual tissue growths in the colon or rectum that can progress into colorectal cancer if left undetected at an early stage. Early and accurate polyp detection during colonoscopy is crucial for effective prevention and treatment. However, manual detection is challenging due to variability in polyp size, shape and texture which can lead to missed or false diagnoses. To address these challenges, we propose a deep learning-based approach to automate polyp detection, capable of identifying even the smallest polyps and aiding early cancer prevention. This research utilizes the Kvasir-SEG dataset, a publicly available collection of annotated polyp images to train and evaluate an advanced detection model. We employed YOLO (You Only Look Once), a state-of-the-art object detection framework and trained its latest version, YOLOv11 on the Kvasir-SEG dataset. The model was enhanced through advanced data preprocessing and hyperparameter tuning, making it suitable for clinical deployment. Our approach achieved excellent results, with an Intersection over Union (IoU) score of 0.9764 and an overall accuracy of 99.00%. Also achieved a balanced precision, recall and F1-score. The detection metrics showed a mean Average Precision (mAP) of 0.9937 at 0.5 IoU threshold and 0.9935 across thresholds from 0.5 to 0.95, indicating robust and reliable performance. The model was comprehensively analyzed with SAM (Segment Anything Model), YOLO-Seg and SAM2. These results demonstrate the effectiveness of our model in accurate and consistent polyp detection. The proposed method can assist clinicians by reducing missed detections and enabling early colorectal cancer diagnosis.

Irritable bowel syndrome (IBS) is a prevalent gastrointestinal disorder characterized by abdominal cramps, pain, bloating, and altered bowel habits, severely affecting the quality of life of the patients. Mebeverine is a common antispasmodic agent that relaxes smooth intestinal muscle, yet its clinical application is limited by systemic side effects. To improve therapeutic targeting and efficacy, we developed a silver nanoparticle (AgNP)-based drug delivery system loaded with mebeverine.
Methods: Drug-loaded AgNPs were synthesized and their effects on smooth muscle contractility were assessed in vitro in the presence of cholinergic inhibitors, selective receptor antagonists, calcium blockers, and neurotransmitters. Anti-inflammatory activity was evaluated via albumin denaturation inhibition.
Results: Mebeverine-loaded AgNPs exhibited distinct effects on contractile response parameters, although they did not directly block cholinergic receptors. In the anti-inflammatory assay, mebeverine demonstrated higher inhibition of albumin denaturation compared to diclofenac, while the nanoparticle formulation retained superior activity relative to diclofenac but was slightly less potent than free mebeverine.
Conclusions: The mebeverine-loaded AgNP system shows promising spasmolytic and anti-inflammatory potential in vitro, supporting its further evaluation as a targeted therapeutic strategy for IBS management.
Acknowledgments:
This study is supported by the Bulgarian Ministry of Education under the National Program “Young Scientists and Postdoctoral Students–2”, Project № MUPD-HF-016.

1. Introduction

Hydrogels are three-dimensional (3D) crosslinked polymeric materials that has been explored for a wide range of applications. In this work, we have fabricated hydrogels with different morphologies for the delivery of a therapeutic, curcumin and an anti-fibrinolytic agent, tranexamic acid.

1. Method

Polyvinyl alcohol-sodium alginate hydrogel films containing various ratios of monomers were prepared by performing three freeze-thaw cycles and ionotropic crosslinking with calcium ions (3%). The swelling capacity, equilibrium water content, water diffusion kinetics, network parameters, and morphology were studied using standard methods. Drug loading and release under various external conditions were studied using UV-Vis absorption spectroscopy.

1. Results

The hydrogels prepared by solvent casting method exhibited a smooth morphology while hydrogels prepared by freeze drawing exhibited a porous and less dense structure. The hydrogels showed good swelling capacity with no disintegration of the material.

1. Conclusions and on-going work

The hydrogels exhibited properties required for wound dressings. With loaded curcumin and tranexamic acid these materials may have potential application as wound dressing bandages. Other studies such as antibacterial features, cell cytotoxicity, mechanical properties are currently in progress.

Materials for Medicine Research Group, School of Medicine, Royal College of Surgeons in Ireland-Medical University of Bahrain, Building No. 2441, Road 2835, Busaiteen Block 228, Kingdom of Bahrain

The N-Queens problem remains a benchmark for evaluating optimization strategies in artificial intelligence due to its combinatorial complexity. This study investigates the performance of four algorithmic approaches—Exhaustive Search (Depth-First Search with Backtracking), Greedy Search (Hill Climbing), Simulated Annealing, and Genetic Algorithm—on varying board sizes (N = 10, 30, 50, 100, 200). The aim is to assess each method's ability to find conflict-free queen placements, and to evaluate their computational efficiency, scalability, and memory usage.
Each algorithm was implemented in Python and tested under identical hardware settings. We designed specific parameter tuning strategies for metaheuristic methods, including cooling schedules for Simulated Annealing and crossover/mutation rates for Genetic Algorithms. Experiments revealed that Exhaustive Search is suitable only for small boards (N ≤ 10), while Greedy Search performs effectively up to N = 50 but struggles with local optima on larger boards. Simulated Annealing offered the best trade-off between time and memory, successfully solving instances up to N = 100. Genetic Algorithm also scaled to N = 100 but with significantly higher memory demands and runtime.
These findings demonstrate that no single algorithm excels universally, highlighting the need for hybrid strategies and parameter optimization to extend scalability. This comparative analysis contributes to understanding algorithm behavior under combinatorial constraints and supports the development of more adaptive optimization techniques for complex real-world problems.