Rangamati Science and Technology University is a public university located in Rangamati, Bangladesh, commonly known as RMSTU. It has faced vocal opposition from local groups such as Parbatya Chattagram Jana Samhati Samiti (PCJSS) because they believe that students, staff, and faculty would come predominantly from outside the Chittagong Hill Tracts, and their settlement in Rangamati would change the character of the region. The first batch of 75 students began classes in November 2015, at a temporary campus set up at Tabalchharhi Shah High School..
This study presents an innovative Rotated Square Split-Ring Resonator-Based (RSSRR) metamaterial design tailored for multiband wireless applications like microwave wireless communication and microwave sensing applications, featuring a 45 & ring;RSSRR. The proposed RS-SRR structure has a very compact electrical length of 0.06 lambda x 0.06 lambda x 0.007 lambda at the lowest resonance frequency with a High Effective Medium Ratio (EMR) value of 15.9. The design was optimized through a comprehensive parametric study where the proposed MTM revealed resonances at 1.3 GHz, 2.3 GHz, 5 GHz, 5.89 GHz, 6.595 GHz, 7.195 GHz, and 8.155 GHz, respectively, indicating its suitability for multiband wireless applications in the L, S, C, and X-bands. The sensing characteristics of the proposed RSSRR-biased sensor model are investigated for both simulation and measurement. The proposed sensor model achieved a high sensitivity of 3.657 and a high Q-factor of 1094.429. An investigation of the 159.5 mm2 area of the array prototype is performed to validate the proposed structure for both microwave wireless communication and sensing applications, where the experimental results strongly agree with the simulation result. The transmission coefficient (S21) behavior is also validated utilizing an equivalent circuit approach. Finally, this study provides a comprehensive analysis of a high EMR-based MTM, demonstrating its potential to significantly enhance performance across multiple frequency bands in microwave wireless communication, as well as in the sensing of solid materials.
The rapid growth of video content on social media platforms, surveillance systems, and educational repositories has made automatic video summarization essential for efficient content consumption and analysis. In this work, we propose a lightweight Transformer-based encoder–decoder model tailored for video summarization. Our approach combines the Transformer’s ability to capture long-range temporal dependencies with architectural optimizations that significantly reduce computational overhead. The model features a compact encoder–decoder structure, a learnable start token, and causal self-attention in the decoder to generate frame-level importance scores autoregressively. Despite having only 2.9 M parameters, the model achieves a state-of-the-art F1-score of approximately 83.22
Tropical forests harbor exceptional biodiversity and provide critical ecological functions through their dynamic ecosystem processes. However, these ecosystems are increasingly threatened by a range of biotic and abiotic disturbances. In the hill forests of Bangladesh, such disturbances have altered ecological processes, yet the mechanisms linking disturbance with soil, plant functional traits, and biodiversity remain poorly understood. This study addresses this knowledge gap using Khadimnagar National Park–a semi-evergreen tropical hill forest–as a case study. The primary objective was to assess the direct and indirect effects of disturbance on soil properties, plant functional traits, and beta diversity. Fifty sample plots were randomly established across the forest. The Principal Component Analysis (PCA) was used to extract key components with the highest loading. These components were then used as composite variables representing the original variables. The analysis was conducted using the Structural Equation Model (SEM) based on composite variables and multiple regression based on original variables. The SEM demonstrated strong model fit (χ2 = 5.34, df = 5, p = 0.376; CFI = 0.991; TLI = 0.975; RMSEA = 0.037; SRMR = 0.068) and revealed that disturbance significantly affected certain functional traits, negatively affecting leaf thickness while positively influencing plant height. Disturbance also increased beta diversity (such as q1) through mechanisms such as niche separation and spatial turnover. Although no significant pathway was observed between disturbance and soil in the SEM, regression analysis revealed that soil organic carbon decreased with increasing bare ground and canopy openness, while cut stems were associated with higher soil bulk density. Overall, these findings suggest that disturbance acts as a strong environmental filter, promoting trait convergence and biotic homogenization, while moderate levels of disturbance may still maintain trait diversity and structural heterogeneity. The SEM further revealed multi-level causal pathways, underscoring the value of trait-based approaches for sustainable disturbance management in small forest systems in tropics, including selective logging, community forestry, and restoration planning.
Adeno-Associated Virus (AAV) serotype classifica- tion is crucial for gene therapy applications and viral vector engineering. This paper presents a novel ensemble deep learning framework that combines multiple feature extraction architectures with advanced data augmentation techniques to achieve superior classification performance. Our proposed methodology integrates ResNet50V2, InceptionV3, and custom convolutional neural networks as feature extractors, followed by ensemble classification using Support Vector Machines with Bayesian optimization. The framework incorporates advanced augmentation strategies including MixUp and CutMix to enhance model generalization. Experimental results on AAV serotype datasets demonstrate exceptional performance with a mean cross-validation accuracy of 98.61%, significantly outperforming traditional single-model approaches. The system achieves perfect classification (100% accuracy) on multiple validation folds, indicating robust and reliable serotype identification capabilities.
Missing data has been a significant challenge in data analysis, reducing the reliability of predictions and disrupting data patterns. In dengue-related datasets, incomplete data has complicated outbreak prediction and public health responses. Traditional imputation techniques often produce biased results, while advanced techniques are computationally intensive and lack transparency. To address this, a novel ensemble weighted average imputation technique combining XGBoost and MiceForest named Gradient-Forest Weighted Average Imputer (GF-WAI) has been proposed. Six additional imputation techniques have been implemented for comparison, and predictive accuracy has been evaluated using RFC, SVC, and NBC. Evaluations have been conducted on both dengue datasets of 1,003 and 10,000 records. The proposed method has outperformed others, achieving an MAE of 641.26 and RMSE of 6743.63 on the small dataset, with significant improvements on the larger dataset, where an MAE of 144.49 and RMSE of 2408.32 have been achieved. An accuracy of 99.90
