Rakibul Hasan Rajib

Rakibul Hasan Rajib

Research Assitatant at CCDS, IUB
Email: rakibul@iub.edu.bd

Publications

You can also find my articles on my Google Scholar profile.

Conference Papers

RGC-BENT: A Novel Dataset for Bent Radio Galaxy Classification

Published in IEEE International Conference on Image Processing, 2025

We present a novel dataset for classifying bent radio active galactic nuclei (AGN), focusing on Narrow-Angle Tail (NAT) and Wide-Angle Tail (WAT) categories. Derived from a recognized radio astronomy survey, the dataset supports detailed classification and benchmarking. We evaluate state-of-the-art deep learning models, with ConvNeXT achieving the highest F1-scores. This resource aims to advance research in bent AGN classification, galaxy cluster dynamics, and galaxy evolution.

BD OPEN LULC MAP: High-resolution Land Use and Land Cover Dataset & Benchmark Results for Developing City — Dhaka, Bangladesh

Published in IEEE International Conference on Image Processing, 2025

We introduce BD Open LULC Map (BOLM)—a high-resolution LULC dataset for Dhaka, Bangladesh—offering pixel-wise annotations across 11 classes over 4,392 km². Ground truth labels were validated by GIS experts. We benchmark segmentation performance using DeepLab V3+ on Bing and Sentinel-2A imagery to support robust modeling and domain adaptation in underrepresented regions.

FedCTTA: A Collaborative Approach to Continual Test-Time Adaptation in Federated Learning

Published in International Joint Conference on Neural Networks, 2025

We propose FedCTTA, a framework for test-time adaptation in federated learning that addresses distribution shifts in dynamic, privacy-sensitive environments. FedCTTA employs teacher-student knowledge distillation, updates only batch normalization layers, and enables similarity-based client collaboration to achieve efficient, continual, and privacy-preserving adaptation. It significantly reduces communication overhead and improves model performance under temporal and spatial heterogeneity, outperforming state-of-the-art methods on benchmark datasets.