Bangladesh University of Business and Technology

Due to their prospective applications in high-efficiency optoelectronic and photovoltaic devices, lead-free halide materials increasingly attract much attention. Despite this interest, basic electrical and optical characteristics of the Cs _3 QR _6 (Q = In, Tl, Ga; R = I, Br, Cl) family remain poorly described. By performing a first-principles study of their electronic structure and optical response, this paper attempts to fill this knowledge gap. Using Density Functional Theory, we investigate the electronic band structures, total and partial density of states, and the optical properties for nine Cs _3 QR _6 compounds. The bandgap values of the materials studied here range within 0.74–1.60 eV when computed with the GGA-PBE functional and within 0.54–2.79 eV with the hybrid HSE06 functional. This is suggested to depend on the electronegativity of the corresponding halide and metal–halide orbital hybridization. The calculation of the total and partial density of states show that p-orbitals dominate near the Fermi level. Optical calculations reveal significant absorption in both the visible and ultraviolet parts of the spectrum. These results stress the possibility of employing Cs _3 QR _6 compounds for further photovoltaic and optoelectronic applications.

Accurate sheep breed classification is critical for modern livestock management, supporting decisions on breeding, productivity, and profitability in the farming industry. Traditional methods based on visual inspection or body measurements are often subjective, time-consuming, and unsuitable for large-scale operations. To address these challenges, this study proposes SheepFormers, a Vision Transformer (ViT)-based framework for automatic sheep breed identification, specifically designed to operate effectively even with small datasets. A balanced dataset of 1,680 sheep face images representing four breeds was preprocessed and used to evaluate multiple transformer variants, including ViT-Google, ViT-MAE, ViT-VAN, BEiT, ViT-ResNet50, and DiT. Through systematic hyperparameter optimization and ablation studies covering epochs, batch size, learning rate, positional encoding, patch size, data augmentation, and 5-fold cross-validation, the ViT-Google model achieved the best performance with 98.21

This article describes a publicly available multimodal Bangla sentiment dataset designed to support research in speech processing, sentiment analysis, and low-resource language modeling. The dataset comprises two synchronized modalities: sentiment-annotated Bangla text and corresponding speech recordings. It contains 1,000 manually curated Bangla sentences evenly distributed across positive and negative sentiment classes, alongside 4,000 aligned audio recordings produced by four native speakers. Each sentence is recorded independently by all speakers to ensure speaker diversity while maintaining consistent textual content. The text component reflects natural, everyday Bangla language usage and is structured to facilitate sentiment classification and linguistic analysis. The audio recordings were collected under controlled yet realistic acoustic conditions using multiple recording devices, introducing natural variability relevant for real-world speech applications. All samples underwent manual quality verification to ensure accurate text-audio alignment and to remove noisy or duplicated recordings. The dataset is suitable for a wide range of applications, including multimodal sentiment classification, sentiment-aware speech recognition, audio-text alignment, and benchmarking of multimodal learning approaches for low-resource languages. Its modular structure allows straightforward extension with additional speakers, dialects, or sentiment categories. By providing aligned textual and speech data for Bangla, this dataset contributes a valuable resource to the research community and supports broader efforts toward linguistic diversity in artificial intelligence.

Single image deraining seeks to restore a clear scene from a single image affected by rain, a task that presents inherent challenges due to the spatially varying, anisotropic, and multiscale characteristics of rain streaks. The current convolutional neural network (CNN)-based approaches are efficient in preserving local texture but poor in long-range dependencies, whereas the Transformer-based approaches are more efficient to model global contextual reasoning at a high cost in computing time. Furthermore, most of the frameworks fail to consider explicit rain-layer models that can be of great structural help when restoring. This paper presents HybridDerainNet, which is a rain-layer guided two-stream CNN-Transformer network that consumes an estimated rain layer as auxiliary input to aid in rain-background separation. The suggested design combines fine-grained spatial preservation using residual convolutional blocks with a small self-attention block to preserve spatial fine details and to effectively combine global information in a simple U-shaped encoder-decoder structure. This design allows local–global representation learning to be balanced and it is computationally efficient. Extensive experiments on Rain1400 bench-mark showed that HybridDerainNet results in 32.78 dB PSNR and 0.93 SSIM in only 0.75 M parameters and 12.8G FLOPs, making a good trade-off between the quality of restoration and model complexity. These findings indicate that explicit rain-aware modeling and hybrid feature extraction is effective in the practical single image deraining.

Deep learning-based automated systems have emerged as powerful tools for medical image analysis. However, existing models often face limitations when applied to gastrointestinal (GI) tract imaging and segmentation because the irregular organ shapes, varying sizes, overlapping regions, and low-contrast boundaries significantly affect the performance and generalizability of current segmentation techniques. To address this challenge, we propose PTransNet, a novel transformer-based segmentation network for GI images. Here, we introduce a novel technique, Contextual Relative Positional Encoding (CRPE), that explicitly embeds relative spatial relationships among features, thereby improving the network’s spatial reasoning in anatomical scenes. We proposed the PTransNet architecture, which combines UNet and a transformer with a novel CRPE for optimized segmentation. The encoder-decoder architecture of UNet enhances model efficiency by capturing contextual features through downsampling and precisely localizing them via symmetric upsampling with skip connections. Using the UW-Madison GI Tract dataset (16,590 images), PTransNet is trained, validated, and tested (80%-10%-10% split) to target the segmentation of the three major GI components (small, large, and stomach), which achieved substantially higher testing performances with a Dice score of 93.49%, an IoU (Intersection-over-Union) of 90.51%, and a specificity of 99.87%. Moreover, PTransNet achieved 10.64% and 13.4% gains in dice and IoU scores, respectively, when using the novel CRPE component, highlighting its contribution to precision diagnosis. The model also outperformed state-of-the-art methods across higher MCC, BM, HD95, NSD, MASD, precision, recall, and F1 scores, demonstrating robust segmentation quality. These advances highlight PTransNet’s strength in handling the challenges of spatial scenarios. Thus, PTransNet has substantial practical implications for the reliable delineation of GI organs in clinical workflows (e.g., radiotherapy planning), thereby contributing to improved healthcare outcomes in the management of GI disease.