Managing waste products generated in the food industry is a pertinent topic in food science. The study aimed to assess the effect of adding shredded black chokeberry (Aronia melanocarpa) pomace in amounts of 0.0%, 0.5%, 1.0%, 2.0%, and 3.0% on the quality of beef burgers subjected to heat treatment, vacuum-packed and stored at refrigeration +4°C for 14 days. On the production day, the heat loss, "shrinkage”, and the content of basic chemical components were measured in the burgers. During storage, pH, shear force, and colour parameters (L*, a*, b*) were measured, organoleptic assessment was carried out, and microbiological quality was evaluated. It was found that using chokeberry pomace as an ingredient in beef burgers affects the quality of these products. With the increase in pomace addition, a significant (p < 0.05) rise in heat losses, greater shrinkage, and an increased fat content were observed. The addition of pomace also resulted in a gradual decrease in the shear force of the burgers. Compared to the control product, burgers with chokeberry pomace were characterised by a significantly (p < 0.05) darker colour, less redness, and less yellowness. In the organoleptic evaluation of all attributes, burgers produced with a lower addition of chokeberry pomace, i.e. 0.5% and 1.0%, received scores similar to the control product. The addition of chokeberry pomace did not cause a deterioration in the microbiological quality of the beef burgers. The amount of chokeberry pomace added to the beef could be 1.0% without negatively affecting the quality of the product.

Sea buckthorn seed oil is a unique plant-based product rich in Omega-3, -6, -7, and -9 fatty acids, along with bioactive antioxidants that support cardiovascular health and promote anti-aging. Traditional extraction methods often involve organic solvents, which may compromise the bioactivity and purity of the oil. Supercritical CO₂ extraction provides a cleaner and more selective alternative. This study aims to model and optimize the process parameters that influence the efficiency and yield of oil extraction from sea buckthorn seeds using supercritical CO₂. A Central Composite Rotatable Design (CCRD) was employed to plan the experiments and evaluate the effects of the parameters. Sea buckthorn seeds were dried and ground before extraction. A supercritical fluid extraction system was utilized. The parameters under investigation included pressure, temperature, and extraction time. Results were processed using Response Surface Methodology (RSM) with Minitab software. Among the parameters studied, pressure exhibited the most significant impact on oil yield, while temperature and time had moderate effects. Interaction effects, particularly pressure–temperature (AB) and pressure–time (AC), significantly influenced the extraction efficiency. The predicted optimal condition for maximum yield was 5075 psi, 70°C, and an extraction time of 10 hours. The study confirms that supercritical CO₂ extraction is effective for obtaining high-quality sea buckthorn seed oil. Pressure was identified as the most critical factor, and the interactions among parameters must be considered for process optimization. These results support future scale-up for the production of functional foods and nutraceuticals.

Bacterial biofilms represent an evolutionary adaptation that enables a shift from a free-living to a cooperative, community-based lifestyle. Within these structures, heterogeneous microbial populations are packaged in a self-produced matrix of extracellular polymeric substances (EPSs), which significantly enhances their resilience, virulence, and persistence across clinical, environmental, and food-related environments. The transition to this hierarchical lifestyle is derived from a multifactorial process centred on Quorum sensing (QS), wherein the exchange of small signalling molecules promotes intercellular communication, and the synchronised expression of genes involved in pathogenicity and antimicrobial resistance. The tolerance exhibited by biofilm-associated cells to environmental stresses and antimicrobials is a non-heritable trait, largely attributed to the protective properties of the EPS matrix, which can inactivate or limit the diffusion of antimicrobial agents. As biofilms continue to pose a global threat, there is a growing demand for innovative and sustainable control strategies. Essential oils (EOs) have gained attention as promising natural antimicrobials due to their multicomponent nature, broad-spectrum activities, and counteraction of bacterial resistance. This review explores the anti-biofilm activity of EOs, with a particular focus on their molecular mechanisms of action against key foodborne pathogens, including Escherichia coli, Listeria monocytogenes, Pseudomonas aeruginosa, Salmonella enterica, and Staphylococcus aureus, by targeting genes involved in QS regulation, motility, adhesion, and virulence.

In response to the escalating cybersecurity threats targeting adolescents, this study proposes an innovative artificial intelligence (AI)-driven framework designed to transform conventional cybersecurity education through personalized and adaptive learning. The system integrates convolutional neural networks (CNN), long short-term memory (LSTM) networks, and natural language processing (NLP) to establish a multimodal behavioral perception mechanism, generating a cybersecurity behavior vector that quantifies vulnerabilities in areas such as phishing susceptibility and privacy protection. A Transformer-based cognitive mapping engine aligns these behavioral features with a structured knowledge graph built from extensive cybersecurity databases, enabling dynamic generation of personalized learning units—including interactive comics, gamified phishing challenges, and AR-based scenarios—optimized via a contextual multi-armed bandit algorithm to enhance long-term retention while minimizing cognitive load. A 12-week randomized controlled trial with 412 middle school students demonstrated that the AI-enabled approach significantly outperformed traditional methods, yielding a 32% increase in knowledge retention, a 2.1-fold improvement in threat detection accuracy, a 28% rise in self-reported security behaviors, and a 41% boost in engagement metrics. The results validate the efficacy of adaptive, data-rich interventions in fostering sustainable cybersecurity habits. The study concludes with policy recommendations for integrating AI-driven personalized education into national cybersecurity strategies, promoting cross-departmental collaboration for resource sharing, and establishing ethical guidelines for equitable and transparent educational AI systems.

Deep learning models for crop classification are crucial for food security (SDG 2) but often fail when deployed in new geographic regions due to domain shift. This is a major barrier in nations like Algeria, which lack large-scale labeled datasets. To address this, we propose an adapted Domain-Adversarial Neural Network (DANN) that effectively transfers knowledge from data-rich European regions to data-scarce Algerian environments for Sentinel-2 imagery. Our methodology is centered on a U-Net segmentation architecture trained with a DANN framework. Our primary contribution is the introduction of a feature-matching loss at the U-Net bottleneck, which forces the model to learn more robust, domain-invariant representations. To address the limited availability of local labeled data, we apply a targeted data augmentation pipeline (including random rotations and scaling) to the small set of labeled Algerian wheat and potato parcels. The proposed model demonstrates significant performance gains. A baseline U-Net trained only on European data achieved 62% accuracy on the Algerian test set. In contrast, our adapted DANN model, trained with only 50% of the available Algerian labels, increased the overall accuracy to 89%. This data-efficient approach yielded high class-specific F1-Scores of 0.93 for wheat and 0.89 for potatoes. This work provides a validated and scalable pathway for developing accurate crop classification systems in regions with limited data.

In recent years, the Mamba architecture and its selective state space models (SSMs) have emerged as next-generation approaches in computer vision, offering linear computational complexity and the ability to efficiently capture long-range dependencies. These properties have attracted significant interest in medical imaging, where computational efficiency and accuracy are critical. This study provides a comprehensive survey of Mamba applications in medical imaging between 2023 and 2025. Representative frameworks, such as VM-UNet, Mamba-UNet, and 2DMamba are examined, focusing on their performance across key tasks including 2D and 3D segmentation, whole-slide pathology image classification, and surgical or endoscopic video analysis. Recent studies indicate that SSMs can achieve performance comparable to, and in many cases surpassing, Transformers in multimodal medical imaging tasks, while substantially reducing memory consumption and computational overhead. These strengths are particularly beneficial for high-resolution and long-sequence applications. Nonetheless, challenges persist in achieving stable optimization, ensuring interpretability, and extending modeling capacity to local features and 3D temporal data. Moreover, the absence of large-scale pre-training resources continues to limit the robustness and generalizability of SSM-based approaches. Overall, Mamba provides a new paradigm for medical image analysis that balances performance with efficiency. Future research should prioritize interpretability, hybrid architecture design, cross-modality generalization, and temporal extensions in 3D imaging to enable smoother translation of this emerging architecture from academic research to clinical practice.

The integration of Generative Artificial Intelligence (GenAI) in English Language Teaching (ELT) is still at an experimental stage, particularly in developing countries where technological access and pedagogical adaptation remain uneven. While global research on GenAI-enhanced language education is expanding, limited attention has been paid to how university-level English teachers in under-resourced Asia-Pacific contexts navigate and adapt to these emerging tools. This study addresses this gap by exploring how Bangladeshi university-level English teachers engage with GenAI to facilitate first-year writing instruction, infusing two theoretical frameworks: the Post Method pedagogy and the TPACK model. Using a qualitative design, structured individual interviews were conducted with ten university English instructors, experimenting with GenAI-assisted teaching for the first time. Thematic findings include: (i) teachers exercised agency by using GenAI as receptive, assistive, and collaborative tools to enhance writing instruction; (ii) GenAI-generated materials were customized to learners’ cognitive, linguistic, and cultural needs; (iii) GenAI supported reflective and adaptive teaching practices; (iv) it promoted learner autonomy and engagement through exemplars and formative feedback; and (v) challenges persisted, including limited GenAI literacy, inadequate institutional policy, unequal resources, and generational resistance. While integration remains experimental, these findings highlight the potential of GenAI to enrich ELT. By offering a “glocal” perspective, the study informs the sustainable, context-sensitive adoption of GenAI in language education, emphasizing pedagogical adaptability, ethical awareness, and institutional support. It underscores the need to bridge the gap between GenAI policies emerging from Global North contexts and their practical, context-sensitive implementation in the Global South, particularly within Asia-Pacific countries.

Parkinson’s Disease (PD) is a progressive neurodegenerative disorder where early detection is essential to slow disease progression and improve patient outcomes. Handwriting analysis, particularly spiral drawings, provides a low-cost, non-invasive biomarker for identifying motor impairments linked to PD. However, existing approaches often depend on multimodal inputs or specialized hardware, limiting scalability and interpretability in clinical use. This study proposes an explainable deep learning framework that detects PD solely from spiral handwriting images.

Using the NewHandPD dataset, a lightweight Convolutional Neural Network (CNN) was trained on grayscale spiral drawings after standard preprocessing steps such as resizing and normalization. The model was evaluated using accuracy, precision, recall, and AUC metrics, and interpretability was ensured through Gradient-Weighted Class Activation Mapping (Grad-CAM), which highlights input regions influencing predictions.

The CNN achieved an overall accuracy of 87%, with a precision of 86.5%, a recall of 87.3%, and an AUC of 0.91 on the test set. Grad-CAM heatmaps confirmed that the network consistently focused on tremor-induced distortions and irregular stroke patterns, aligning with clinically relevant features of PD. This interpretability bridges the gap between deep learning performance and clinical trust, addressing the common “black box” limitation of AI systems in healthcare.

The results demonstrate that handwriting-based biometrics can serve as an effective, explainable, and deployable tool for early PD screening. This work provides a foundation for integrating transparent AI-driven diagnostics into clinical workflows and expanding research toward multimodal approaches in neurodegenerative disease detection.

Wildfires have emerged as one of the most pressing environmental hazards, fueled by climate change, deforestation, and human activity. Early detection is critical to reducing damage to ecosystems and human settlements, yet traditional monitoring methods such as ground sensors or manual observation are slow and limited in scope. Recent advances in deep learning have shown promise, but many approaches rely on heavy segmentation or object detection models that are unsuitable for real-time or resource-constrained environments.

This study proposes a lightweight wildfire detection framework using EfficientNetV2-S with transfer learning for binary image classification. A publicly available satellite image dataset was preprocessed through resizing, normalization, and train–test splitting. Transfer learning was applied by freezing pretrained ImageNet weights and fine-tuning the classifier head for two categories: fire and no-fire. The model was trained using the AdamW optimizer with cross-entropy loss and evaluated through accuracy, precision, recall, F1-score, and confusion matrices.

The system achieved 97.54% validation accuracy and 92.65% test accuracy, with balanced precision and recall across both classes. Robustness testing on real-world satellite images confirmed strong generalization, while inference times of 15–20 ms per image demonstrated real-time viability. Unlike heavier segmentation-based pipelines, this lightweight model can be deployed on drones, edge devices, and early-warning platforms.

In conclusion, EfficientNetV2-S provides an efficient, accurate, and scalable solution for wildfire detection, offering a deployable alternative to computationally intensive models and supporting rapid-response systems for disaster prevention.

Pneumonia remains a leading cause of morbidity and mortality worldwide, particularly among children and the elderly. Early and accurate diagnosis is vital to improving patient outcomes; however, manual chest X-ray interpretation requires specialized expertise and is prone to subjectivity. Deep learning offers a promising solution by automating diagnosis while reducing diagnostic delays and improving accessibility. This study develops and evaluates a lightweight Convolutional Neural Network (CNN) model for binary classification of chest X-ray images into “Normal” and “Pneumonia” categories.

The publicly available Kaggle Chest X-Ray Pneumonia dataset, comprising 5,863 pediatric radiographs, was used for training, validation, and testing. Images were preprocessed through resizing, normalization, and data augmentation techniques including flipping and rotation to enhance model generalization. The CNN architecture included three convolutional blocks followed by dense layers, dropout regularization, and a final sigmoid classifier. Training was conducted for 15 epochs with the Adam optimizer, and performance was assessed using accuracy, precision, recall, and F1-score.

Results demonstrate a test accuracy of 76.6%, with precision of 90% for pneumonia cases and recall of 94% for normal cases. The model showed strong diagnostic capability for normal scans but occasionally misclassified subtle pneumonia features, as confirmed by means of qualitative error analysis. Despite these limitations, the CNN achieved balanced performance with reduced computational complexity, making it suitable for deployment in resource-limited settings.

In conclusion, this study highlights the potential of efficient CNN architectures for supporting pneumonia diagnosis, offering a scalable and interpretable tool for clinical decision support and preliminary screening.