Plastic pollution is a key environmental issue affecting communities worldwide, with millions of plastic bottles each day ending up in landfills for the absence of efficient recycling infrastructure. This project created a social enterprise that transforms common plastic waste into stylish fashion items, such as household, fashion, community industrial accessories etc. The initiative uses digital tools to collect plastic efficiently, combining circular economy principles with innovative upcycling and eco-friendly marketing that resonates with environmentally conscious consumers. Consumer demand for green clothing shall be determined through questionnaires, while major-influencing actors in a local environment shall provide opinions regarding challenges and prospects in operation through interview schedules. It also highlights empowerment of female citizens, unemployed young and low-income citizens through engagement in the production corridor for attainable means of livelihood. Its projected contribution shall witness a designated measure in minimization of plastic wastes estimated at 200–300 metric tonnes annually in pilot locations and an addition of 20–30 employment for local citizens. The project shows promise for setting up community micro-factories, integrating digital tools into the supply chain, and creating a model that can be replicated across the country. Providing a reproduceable, scalable template, the framework establishes a means for transforming environmental challenges to business opportunities for the provision of social, economic, and ecological values. It generates additional knowledge in the subject of sustainable entrepreneurship such as in utilizing the strategy of circular economy and social innovations, offering international relevancy for mitigating plastic pollution through enterprise.
This study is motivated by the need to better understand and predict the conditions under which electromagnetic exposure induces a flight response in biological subjects, which is important for assessing safety thresholds and guiding the design of exposure tests. The aim of this work is to investigate the occurrence of the flight reflex when a subject is exposed to an electromagnetic beam moving relative to the body. A heat-induced flight model is developed based on the physical processes of beam energy absorption, heat diffusion across the skin, activation of thermal nociceptors, and the initiation of a flight response once the activated skin volume reaches a critical threshold. In the present mathematical model, the governing system is reduced to a normalized formulation in which the temperature field reaches a steady state in a coordinate system moving with the beam. A key parameter in the heating process is the active heating duration of the beam defined as the time it takes for the moving beam to traverse one beam radius. Within this framework, the occurrence of flight is analyzed in the two-dimensional space of physical beam radius and power density at various values of beam active heating duration, which contains the effect of beam moving velocity. An accurate empirical formula along with a simple scaling law is derived to characterize the boundary of the flight-inducing region. The results show that the scaling law reliably predicts the effects of varying beam radius and velocity, even without detailed model parameters. This provides a practical tool for forecasting flight-inducing exposure conditions and supports safety assessment and system design for moving electromagnetic sources.
Ballistic protection structures have a long history of effectively countering various threats. However, a persistent challenge remains in enhancing their ballistic performance while minimizing weight. To achieve optimized structures, understanding the penetration behavior of constituent materials and energy absorption mechanisms is crucial. The quasi-static punch shear test (QS-PST) is commonly employed, particularly in composite materials, despite its significantly lower strain rates compared to ballistic events. This test serves as a valuable tool for gaining insights into energy absorption and failure mechanisms during high-speed impacts. Moreover, the outcomes of QS-PST have contributed to the development of predictive models for terminal ballistics and dynamic energy resistance in protective structures. This, in turn, provides designers with valuable insights into the ballistic resistance of structures, minimizing the need for expensive ballistic testing. This article presents a thorough review of research conducted using the QS-PST to characterize the penetration behavior of composite structures. It delves into the key factors influencing the test and offers a comprehensive methodology overview for analyzing penetration in materials commonly utilized in composite bullet-resistant structures. The literature review uncovers various studies leveraging the QS-PST to assess the penetration resistance of diverse targets. Additionally, it explores how the results can be correlated with high-speed impact events, shedding light on the energy absorbed through damage mechanisms observed in the materials.
Skin disease are major global public health concern, and their accurate diagnosis usually demands comprehensive clinical examination, In the wake of recent interest in the development of automated detection methods for skin diseases, several artificial intelligence based techniques have been proposed; most of these techniques, however, are based on pre-trained deep learning architectures, which are computationally expensive to train, thus limiting their real-world applicability in resource-constrained clinical environments. Addressing such limitations, this work proposes a lightweight attention-based convolutional neural network optimized using a Genetic Algorithm (GA) for multi-class skin disease classification. Accordingly, the proposed framework embeds an attention mechanism to enhance lesion-relevant regions of interest while using GA-based hyperparameter optimization to improve the robustness, generalization, and classification performance of the network. The model has been tested on the Kaggle Monkeypox 2022 Remastered dataset, comprising 847 images belonging to six classes of skin diseases. Performance was evaluated using 5-fold cross-validation to ensure robustness and reliability of the reported results. The experimental outcome shows that the proposed GA-CNN-Attn framework yields an accuracy of 98.18
The majority of services are now delivered to end users, clients, or citizens through Information and Communication Technology (ICT), as internet access has expanded beyond urban centers to reach rural and remote areas. This widespread connectivity makes systems and services—whether offered by government, businesses, or other organizations—easily accessible via ICT-based platforms. In order to serve their clients, these services use distributed computing, shared resources, and networked services in both central and multi-server-based architectures and the biggest challenge for these systems are security which is a must for deployment because of the system’s openness. According to the stan- dard ISO security model, one of the most important security services required for every ICT-based system—and the same is true for systems based on multi-server architecture—is authentication. This work develops an authentication protocol for a system based on a multi-server architecture that does not rely on a central verification table and has the ability to store authentication data in distributed manner. The suggested protocol offers the ability to share parameters across all communication tiers in order to determine the session key for subsequent communication, in addition to this authentication service. The proposed protocol’s goal is to combine key agreement and authentication into a single protocol with distributed storage functionality. The proposed protocol integrates authentication and session key agreement in a unified framework, facilitating seamless access to services through a single registration process. It is designed with applicability to systems like Indian e-governance, it address security challenges arising from system openness. It is rigorously evaluated against 21 security goals. The cost of computing for any of the security protocols is in general based on used hashing, and this protocol costs 22Thash and the total 496 bytes for computation and communication respectively.
Various living and non-living factors can significantly decrease the rice yield. Existing approaches to deep learning often fail to consider the influence of environmental factors on disease and severity classification. A multimodal deep learning model for predicting disease (10 classes) and severity (3 levels) of rice leaves continuously through a combination of visual disease features and environmental factors. The framework uses a dual-branch architecture: (i) a dual Convolution Neural Network (CNN)-based visual encoder to extract the features of the lesions, and (ii) a Multilayer Perceptron (MLP)-based environmental encoder to model the environmental conditions in the field. Fused images using cross-attention module, followed by dual head outputs for disease and severity prediction. Gradient-weighted Class Activation Mapping (Grad-CAM) visualization is used for the interpretation of lesion-focused decisions. The model is trained using multi-task loss and tested using five-fold cross-validation. This model obtained 97.9
Our proposed architecture, AutoSortBin, is based on the Cyber Physical System (CPS) and Internet of Things (IoT) technology which emphasizes the importance of proper waste disposal and segregation. To tackle the problem of waste segregation and management, AutoSortBin automatically classifies and sorts the waste into six broad categories namely metal, paper, plastic, glass, organic, and e-waste. The technical idea behind this segregation is implemented by the DenseNet-121 model for transfer learning integrated into the Wokwi simulator using ThingSpeak. The proposed framework takes an image input through a camera and identifies the waste as one of the main categories. The output from the waste identification model serves as an input for the IoT-based circuit and it opens the corresponding waste bin lid via servo motors. It also uses ultrasonic distance sensor to monitor the storage level in the waste bins. The proposed framework demonstrates an automation system that alerts the authorities to empty a bin whenever a bin is full by sending emails, to solve the problem of waste management efficiently thus contributing to environmental sustainability. The proposed framework has high potential for scalability and integration with different CPS and IoT platforms for enhanced performance and features. The setup exhibited an exceptional accuracy of 94.63
In recent years, due to the massive amount of information on the web, the rumor detection on social media presents a significant challenge owing to online content’s noisy, dynamic, and often adversarial nature. This work introduces a model that leverages Graph Attention Networks (GAT) enhanced with adversarial and contrastive learning to improve rumor classification performance. Experimental results on the X Dataset (formerly Twitter) demonstrate that our integrated GAT+ADV+CL model achieves satisfactory performance across multiple classification evaluation metrics, while maintaining a relatively simple architecture compared to other recent complex graph-based approaches for rumor detection. These findings highlight the effectiveness of combining robustness and representation learning in tackling the challenge of misinformation detection.
Traumatic Brain Injury, which is also known as TBI, is a serious global health concern, with neuroimaging playing a vital role in its diagnosis and prognosis. Skull stripping refers to the technique of isolating the brain region by eliminating non-brain structures from neuroimaging data. It is an essential preprocessing step in neuroimaging analysis to improve lesion detection, feature extraction, and tissue classification. This study presents an efficient skull stripping method, ThresContCT_TBI, specifically designed for computed tomography (CT) images of TBI patients. ThresContCT_TBI integrates intensity thresholding, morphological processing, contour filtering, and distance transforms to achieve robust brain extraction while addressing challenges such as traumatic lesions and anatomical complexities. The effectiveness of the approach was evaluated using segmentation metrics, including accuracy (98.68
Valvular Heart Disease (VHD), caused by malfunctioning heart valves, poses significant diagnostic challenges due to the complexity of heart sound patterns and variability in clinical presentations. Traditional auscultation methods are subjective, and existing automated models often function as black boxes, offering limited insight into the reasoning behind predictions. Therefore, there is a pressing need for accurate, interpretable, and multi-class diagnostic tools to aid clinicians in early and reliable detection of VHD using phonocardiogram (PCG) signals. X-CBNet, a convolutional neural network (CNN) followed by a bidirectional long short-term memory (Bi-LSTM) network framework, is employed in this research to harness deep learning’s capability to achieve high diagnostic accuracy while ensuring interpretability through explainable AI methods. Melspectrograms are used to capture essential features of the phonocardiograms. The proposed model is designed as a five-class classifier distinguishing between aortic stenosis, mitral stenosis, mitral regurgitation, mitral valve prolapse, and normal heart sounds. Gradient-weighted class activation mapping (Grad-CAM) is utilized for explainability, generating heatmaps that visualize model decision-making. The proposed X-CBNet model achieved an overall accuracy of 99.15
As per the report by the National Institutes of Health (NIH), it has been found that pneumonia has 16 times and 10 times more reported cases than cancer and HIV-AIDS, respectively. X-ray scans are used by radiologists to diagnose pneumonia. Early disease diagnosis may prevent disease progression and improve patients’ lifespan. This work proposes deep learning-based stacked ensemble models by using SqueezeNet and VGG16 feature extractors integrated with hybrid trio stacked classification models. The proposed models have been implemented on a pooled dataset collected from three different sources with variations in the data to extract the most optimal features. The hyperparameters of these models have been configured with varied values to achieve higher outcomes for better disease detection. The performance evaluation of the VGG16-based ensembled model exhibits an accuracy of 98.01
Auto text summarization (ATS) revolutionized almost every area from social media to healthcare by giving the main information or summary of the document in a quick span of time. Extractive and abstractive summarization techniques are the pinnacle of advancement in ATS. Abstractive text summarization generates fluent and human-like summaries as new text, but this summary suffers from the problem of intrinsic or extrinsic hallucination. Filling this research gap is crucial because it undermines the factuality, faithfulness, and reliability of the generated summary by introducing plausible text that does not exist in the original text. While extractive techniques are fast and faithful, they lack fluency and coherence because the text is taken directly from the source material. To address the research gap of hallucination in generated summaries, this paper proposed an improved dual-stage hybrid approach by combining the functionalities of an extractive graph-based model and a transformer-based abstractive ATS model. The proposed approach is an improved approach for generating more factual summaries using hybridization. The proposed dual approach first extracts the important information from the main text and then processes this information using deep generative architecture like transformers. By ensuring the strength and semantics of selected content, the proposed approach generates a consistent summary. Hallucinations are detected based on the extra entities present in the generated summary with respect to the source text using the entity-level verification method named “entity extraction.” Entity matching is performed to determine the extra, missing entity. Further, to evaluate the performance of the proposed approach, it is tested for factuality and compared with the base model using precision, recall, and the F-measure of the ROUGE score on the CNN/DailyMail and XSum datasets. The proposed model has the highest recall 0.42 and 0.31 on CNN/DailyMail and XSum and the highest similarity score 0.87 as compared to the base model. The model is also evaluated on performance metrics like computational cost, latency, and model size. Hallucination is detected and reduced in generated summaries on the basis of extraneous, missing entities, and similarity scores with respect to source data using the proposed strategy. A statistical test is performed to check the superiority of the proposed model over the base model.
Pneumonia continues to be one of the global causes of morbidity and mortality, necessitating early and proper diagnosis using chest X-rays (CXRs). Deep learning networks, specifically convolutional neural networks (CNNs), have shown excellent results in identifying pneumonia. But their performance is normally hampered by dataset imbalance, feature redundancy, and local receptive fields, reducing generalizability. To overcome such challenges, the study introduces Vi-GeN 1.0, a GAN-augmented Vision Transformer (ViT) pipeline specifically architected for pneumonia classification robustness. Our method uses Wasserstein GAN with Gradient Penalty (WGAN-GP) to synthesize high-quality CXRs, improving dataset diversity and feature learning. The ViT classifier, trained on the real and synthetic data, learns global contextual representations, resulting in better classification performance. Vi-GeN 1.0 model was tested on the Kaggle Chest X-ray Pneumonia dataset, showing 97.3
The paper introduces a novel hybrid Feedback-Based Quantum Optimization (FBQO) framework that integrates reinforcement learning (RL) for adaptive parameter control and Kalman filters for noise mitigation to enhance quantum optimization on noisy intermediate-scale quantum (NISQ) devices. Unlike traditional quantum Approximate optimization algorithm (QAOA), the proposed method dynamically tunes parameters through learned policies and statistically filtered feedback, enabling faster convergence and greater noise resilience. Key contributions include the formulation of quantum optimization as a Markov Decision Process, integration of deep quantum networks (DQNs) for adaptive control, and the use of Kalman filtering for robust state estimation. Experimental results on the Max-Cut problem demonstrate superior performance in convergence rate, stability, and optimization accuracy over existing techniques. The main finding is that RL-FBQO achieves convergence in 10–20 iterations, compared to 20–30 for FBQO and 40–50 for QAOA, with superior stability.
As a result of the traditional credit assessment method’s reliance on the financial analyst’s experience and the fact that the majority of the information comes from the debtor, lending institutions are currently faced with a significant issue in predicting credit risk. Machine learning algorithms were therefore developed to ascertain whether a borrower was qualified to seek a loan and committed to paying installments. In this research, a potential effort is made to apply feature engineering techniques, such as missing data correction, data transformation, feature selection, resampling unbalanced target class, and also comparing prediction accuracy of the most prominent ensemble machine learning models (i.e. XGBoostC, LGBMC, CatBoostC) that have special interest in the classification of credit scoring by using Berhan bank’s ten years credit dataset. The dataset has over 47,000 loan records and 16 features with personal, loan, and collateral information. To split the dataset for training and testing, the StratifiedKFold algorithm is used with tenfold cross-validation. Confusion matrix and AUC-ROC metrics are used to evaluate the performance of machine learning models. Finally, the outcome demonstrates that the outperformer ensemble model, CatBoost Classifier, has superior prediction accuracy in terms of AUC = 87
As a major global health challenge, the progressive cognitive decline and memory loss associated with Alzheimer's disease (AD) significantly impact individuals' quality of life. Due to the lack of a definitive cure, early and precise diagnosis remains crucial for implementing effective intervention and management strategies. In this study, we introduce a pioneering approach utilizing Transformer based ResLadderNet for AD classification, achieving an exceptional accuracy of 98
Direct speech-to-speech translation (S2ST) is an important tool for bridging communication gaps. Direct S2ST translates speech from one language to another without relying on intermediate text, making it particularly useful for languages primarily spoken rather than written. However, the performance of Direct S2ST models on low-resource languages remains limited due to the scarcity or complete absence of parallel speech data required for training. Pretraining and finetuning are widely used techniques to leverage unsupervised speech data to improve model performance. In this work, we employ a cluster-aided, cross-contrastive self-supervised learning (SSL)-based speech representation model as the pre-trained encoder, combined with a multilingual BART (mBART) decoder. The resulting finetuned model outperforms a baseline that uses a contrastive-loss-based SSL model as the encoder. The proposed models improve the BLEU score by 4.14 → English and 8.2 → Hindi compared to their respective baseline models. To train the model for English-to-Hindi, we trained a unit-vocoder on speech quantized using ensemble clustering instead of standard clustering. The resulting unit-vocoder outperformed the one trained on speech quantized using standard k-means for all evaluation metrics.
To reduce the considerable risk of infection associated with traditional open surgery, robot-assisted laparoscopic surgery is employed. However, during the robot-assisted surgery, smoke is produced, which lowers image visibility; removing smoke from images is comes under image dehazing. Several image dehazing methods based on different architectures have been created to enhance damaged images. We presented an attention-based Y-Net (AY-Net) to help in the surgery by dehazing smoked laparoscopic surgical images. The AY-Net architecture consists of an encoder, bottleneck, decoder, and mixed decoder, with skip connections between the encoder, decoder, and mixed decoder. The encoder contains encoder layers, and the decoder contains decoder layers, both of which have CNN block sublayers and the mixed decoder has a mixed layer has its sublayers. The proposed Attention Y-Net performed well, on Outdoor, Dense-Haze, and DeSmoke-LAP dataset when compared to other current approaches for image dehazing, both quantitatively and visually.
This study introduces an innovative coverless steganography method aimed at enhancing the security and efficiency of covert communication through digital images. The proposed method utilizes a pre-trained Vision Transformer (ViT) model to extract features and then divides the extracted features into non-overlapping blocks. The average coefficient for each feature block was calculated for hash sequence generation. Subsequently, a binary hash sequence is generated from the mean coefficients of feature blocks. Utilizing inverted indexing, the method dynamically manages hash sequences and their corresponding images and ensures the accurate extraction of confidential data. The results indicate that the proposed coverless information method significantly enhances the imperceptibility and security of hidden data compared to traditional approaches, rendering it a valuable technique for covert communication. The proposed method demonstrates exceptional performance in concealing confidential information within the images. It can embed up to 20 bits per image while maintaining near-perfect accuracy against numerous non-geometric attacks. For geometric attacks, such as rotation, the method achieved an accuracy of over 80
Glaucoma is a progressive eye disease that can lead to irreversible vision loss if not detected and treated early. Elevated intraocular pressure, caused by impaired aqueous humor flow, damages the optic nerve and associated visual pathways. Routine eye examinations are essential for early diagnosis. Deep learning (DL) techniques have shown great potential in automating glaucoma detection from retinal fundus images with high accuracy and minimal expert intervention. This review presents a comprehensive analysis of recent DL-based approaches for glaucoma detection, covering image preprocessing, optic disc and cup segmentation, imaging modalities, benchmark datasets including ACRIMA, Drishti-GS1 and RIMONE, and evaluation metrics. The review further presents key findings and recommendations, and concludes by highlighting current challenges and future research directions to improve the clinical applicability and diagnostic performance of DL-based systems. Notably, several DL studies report over 98
