Developed a novel BMFCNet architecture integrating blended multi-level feature extraction for robust detection of Major Depressive Disorder (MDD) from EEG signals. Introduced a Residual-Inception module to effectively capture both low-level (LL) and high-level (HL) discriminative EEG features, enhancing representational capacity. Developed a Constraint Fusion mechanism for adaptive weighting and fusion of LL and HL features, improving feature integration and classification performance. Addressed subjectivity in MDD diagnosis by providing an automated, EEG-based framework that enhances accuracy, reliability, and clinical applicability. Validated the proposed model on benchmark datasets, demonstrating superior performance compared to 16 state-of-the-art methods in terms of accuracy and efficiency.

• Veracity-aware fusion combines CLIP, SBERT, and NLI to capture image–text incongruity for sarcasm understanding. • Ordinal sarcasm modeling predicts intensity levels (Not, Mild, High) instead of simple binary classification. • Dual-view learning uses original and masked-text memes to reduce template bias and improve generalization. • CLIP + transformer architecture enables effective cross-modal interaction between image and text features. • Template-grouped 5-fold validation prevents data leakage and ensures realistic model evaluation. • Multi-loss training strategy (contrastive, incongruity, ordinal) improves feature alignment and detection performance.

• The system utilises Wi-Fi signal strength mapping to follow a designated operator, employing unique identification to ensure precise tracking in crowded environments. • Equipped with ultrasonic sensors for real-time person tracking and Bluetooth-enabled remote control for manual operation, the trolley maintains optimal distance and adapts to changes in the operator's speed. • Additionally, a load cell system ensures safe load management and alerts the user above the threshold limit of weight.

• Converts solar irradiance into chemical energy to produce hydrogen via water electrolysis, enabling renewable fuel generation. • Employs a hybrid dual-fuel engine (hydrogen–diesel blend) optimized for irrigation pumping, enhancing combustion efficiency and operational reliability. • Significantly decreases reliance on conventional fossil fuels while mitigating greenhouse gas emissions and environmental impact. • Delivers a cost-efficient, low-carbon solution for agricultural irrigation with reduced long-term operating expenses. • Demonstrates high scalability and adaptability across diverse farm sizes, irrigation demands, and geographic conditions. • Achieves improved overall energy efficiency relative to conventional diesel-powered irrigation systems through hybridization.

• Dual-mode operation (Heating + Cooling) in a single portable bottle • Based on thermoelectric (Peltier) effect • Manual DPDT switch-based control (no microcontroller required) • Rechargeable battery-powered system • Integrated heat sink + cooling fan assembly • Compact, portable, and self-contained design • Eliminates dependency on: Refrigerator, Kettle • Low complexity, high reliability system • Suitable for outdoor, travel, medical, and sports use

TinyEyeNet introduces a lightweight CNN for accurate anterior segment eye disease classification Designed for high performance with low computational cost, ideal for real-world deployment Trained and validated on a custom-curated clinical eye image dataset Achieves strong diagnostic accuracy, outperforming conventional deep models Suitable for resource-constrained and portable ophthalmic screening systems Enables faster, scalable, and accessible eye disease detection

• The apparatus leverages sustainable photovoltaic energy, offering a viable alternative to traditional grid-dependent or fossil-fuel-reliant machinery in off-grid rural locations. • The design incorporates a five-tier screen configuration with diverse apertures, allowing for precise sorting and classification of harvests by dimensions and mass. • A specialised rotary transmission unit powers the oscillatory motion of the sieves, replacing the need for arduous manual labour. • Performance evaluations conducted on wheat, garbanzo beans, and mustard seeds confirmed the unit's effectiveness across various grain types and contamination levels. • Technology reached a purification rate exceeding 98%, while simultaneously lowering overhead expenses and physical strain for farmers. • Tests highlighted the machine’s structural durability and its capacity for prolonged use without thermal issues or mechanical failure.

• The study investigates both mechanical and thermal performance of hollow core slabs, focusing on compressive, flexural strength, fire resistance, and heat insulation characteristics. • Results show that hollow core slabs provide better thermal insulation, with approximately 6°C lower temperature and improved heat efficiency compared to conventional slabs. • The U-value of hollow core slab is lower, indicating better insulating performance, and overall heat efficiency is significantly higher (~80%) than normal slabs. • Although flexural strength is slightly lower (˜10% less) than conventional slabs, hollow core slabs are found to be economical, durable, and suitable for hot and humid climates.

• Dual-mode mobility integrating aerial and ground operation in a single platform • In-place transformation without forward motion, unlike conventional hybrid robots • Body-lift mechanism enabling smooth and controlled mode switching • Linear actuator-based transformation avoiding wheel scraping and surface friction • Reduced mechanical wear compared to servo-driven wheel rotation systems • Improved terrain adaptability by maintaining natural wheel orientation during transition • Scalable actuator design with potential for speed optimization and performance enhancement

• Spider robot design enables stable movement on sand and rough desert terrain. • Equipped with sensors to identify and locate water sources accurately. • Allows operators to monitor the robot’s status and control its operations remotely via a secure wireless connection. • Integrates with cloud computing platforms for real-time data transmission, analysis, and visualization.

This work presents RCAN-RS, an enhanced Residual Channel Attention Network designed for remote sensing image super-resolution. Building upon the RCAN framework, the proposed model incorporates domain-specific improvements, including dual-pooling channel attention, spectral attention, and an edge enhancement module to better preserve spectral information, structural details, and sharp boundaries in satellite imagery. Trained and evaluated on the DOTA dataset with a 2× super-resolution setting, the model achieved strong performance with a PSNR of 34.42 dB, SSIM of 0.9398, EPI of 0.9524, SAM of 0.9830, ERGAS of 6.68, and UQI of 0.9846. These results demonstrate the effectiveness of attention-guided and edge-aware mechanisms for generating high-quality, detail-preserving super-resolved satellite images.

This work proposes an ECA-ResNet-based facial emotion recognition system as a supplementary tool for mental health assessment. The model integrates enhanced channel-spatial attention, residual learning, mixed precision training, and a cosine annealing learning rate scheduler to improve feature extraction, training efficiency, and convergence. Evaluated on the OAHEGA Emotion Recognition Dataset containing six emotional classes (Angry, Surprise, Sad, Ahegao, Happy, and Neutral), the model achieved an overall accuracy of 81%. The attention mechanism enables the network to focus on emotionally significant facial regions, while residual connections enhance feature learning and reduce vanishing gradient issues. The results demonstrate the potential of the proposed framework as a reliable foundation for AI-driven mental health support systems, enabling early emotional screening and more accessible digital psychological care.

1. Multilingual clinical text processing: The framework automatically identifies language, performs transformer-based translation, and normalizes medical semantics to preserve diagnostic meaning. 2. Multimodal feature fusion: It combines clinical text features from FastText and ClinicalBERT with multi-view ultrasound visual features extracted using SwinV2/DINOv2. 3. Automated report generation: A cross-attention fusion module aligns text and image information, followed by a GPT-V2 decoder to generate clinically meaningful medical reports.

This work presents an attention-based CNN–Transformer model for classifying ultrasonic plant stress sounds in noisy agricultural environments. By processing raw 1D audio waveforms, the model preserves fine temporal features, while CNN layers extract local patterns and the Transformer captures long-range dependencies through self-attention. Positional encoding and weighted cross-entropy loss further improve temporal understanding and address class imbalance. The proposed approach outperforms conventional machine learning and CNN-based methods, achieving 79.5% classification accuracy, 82% species-level discrimination accuracy, and 99.5% accuracy in binary stress detection (stressed vs. control). These results highlight its potential for reliable, non-invasive, and real-time plant stress monitoring in precision agriculture.

• FarmGuard AI introduces a deep learning–based system for early plant disease detection using leaf images. • Uses CNN models for accurate and real-time crop health analysis. • Scalable and efficient, ensuring high performance with low computational cost. • Trained on diverse plant datasets for reliable and robust predictions. • Provides actionable insights, including disease identification and preventive measures. • Integrates weather-based alerts to improve decision-making. • Reduces pesticide usage through precise and timely intervention. • Promotes sustainable and precision agriculture practices.