IEEE MIT URTC 2025Full Paper · 1st author

HybiNet: Multi-Model Feature Fusion with Deep Explainability for Chest X-ray Diagnosis

A. F. H. Dhrubo, M. A. Rusho, M. K. Jahan, M. A. Qayum

A block-level fusion ensemble that injects intermediate features from EfficientNetB0, DenseNet121, and Xception into customized SqueezeNet Fire Modules — reaching heavyweight accuracy with far fewer parameters: 95.44% validation accuracy at 2.83M parameters, a 62.6% reduction versus DenseNet121, with strong interpretability. HybiNet v2 extends the design to object detection...

📃 Paper 🛠️ Code

IEEE IPAS 2025Full Paper · contributor

🏵️ Best Session Paper Award

Multi-Model AI-Driven Satellite Image Segmentation for Comprehensive Urbanization and Environmental Degradation Analysis in Bangladesh

M. Rezwan-ul-alam, A. F. H. Dhrubo, et al.

A large-scale annotated dataset and temporal-analysis framework covering buildings, trees, water bodies, tin shades, and farmlands across a decade of satellite imagery — quantifying land-use transformation and environmental degradation in Bangladesh. I built the annotation workflow and the temporal analysis pipeline...

📃 Paper

ICAIIC 2025Full Paper · 2nd author

Early Detection of Lung Cancer Using Pyramid Vision Transformer (PVTv2): A Comparative Analysis of Deep Learning Models

M. K. Jahan, A. F. H. Dhrubo, M. A. Rusho, A. R. Chowdhury, F. Sharmin, M. A. Qayum

Systematic benchmarking of transformer-based vision architectures with ablation studies, identifying PVTv2 as the strongest performer for lung-cancer detection. The paper grew out of broader work on which Vision Transformer variants generalize best in resource-constrained medical imaging...

📃 Paper

ICESA 2024Full Paper · 1st author

Harnessing Technology for Climate Resilience: A New Paradigm in Urban Cooling through Eco-friendly Cooler

A. F. H. Dhrubo, M. A. Qayum, et al.

A sustainable alternative to conventional air conditioning for Bangladesh's urban-heat crisis. I contributed an AI-driven control algorithm based on ASHRAE 55 (PMV/PPD) comfort modeling and helped design a hybrid architecture combining MOF-based desiccant dehumidification with evaporative cooling. The prototype delivers cooling comparable to a 1-ton AC while consuming 67–75% less energy at roughly one-third the cost...

📃 Paper

MMAR 2024Full Paper · 1st author

IoT Towers of Resilience: A Revolutionary Approach to Weathering Climate Extremes and Enhancing Sustainability in Vulnerable Ecosystems

A. F. H. Dhrubo, S. Siddique, M. A. Qayum

A priority-based sensor-fusion mechanism using flame, temperature, and humidity inputs to classify fire severity and trigger appropriate responses. The system includes AtmoBalancer, an urban-heat mitigation module via automated mist deployment, and a three-tower distributed network of solar-powered nodes connected to a Raspberry Pi Pico hub for real-time Firebase aggregation and predictive climate modeling...

📃 Paper

ACL 2026Under review · 1st author

LDCS: Legal Document Classification & Summarization with Kolmogorov–Arnold Networks

A. F. H. Dhrubo, et al.

A deep-learning approach to legal document classification and summarization built to tackle domain-specific language, long-term dependencies, and class imbalance. A KAN block is added to the model to better process complex legal texts; extensive experiments on a Bangladesh legal dataset demonstrate substantial improvements over standard ML baselines such as SVM and logistic regression...

Frontiers in Medicine 2026Journal Article

CerevianNet: Parameter Efficient Multi-Class Brain Tumor Classification Using Custom Lightweight CNN

M. K. Jahan, A. A. Shafi, M. A. Rusho, M. S. Hussain, A. F. H. Dhrubo

📃 Paper

PLOS ONE 2026Journal Article

PlantaNet and PlantaNet-Lite: Efficient and Explainable Multi-Crop Plant Disease Classification via Transformer Benchmarking and Custom Lightweight CNNs

M. S. H. Zidan, A. Imran, M. K. Mia, M. Hussain, A. F. H. Dhrubo, M. A. Qayum

📃 Paper

IEEE IPAS 2025Full Paper · contributor

Optimizing Multimodal Transformers for Medical Image Captioning: Enhancing Automated Descriptions via AI Systems

M. Arman, A. F. H. Dhrubo, et al.

📃 Paper

ACM Computing SurveysUnder review · 1st author

Scalable and Secure Blockchain-Based Storage Systems with AI Integration: A Survey of Architectures, Incentives, and Access Mechanisms

A. F. H. Dhrubo, et al.

Knowledge-Based Systems (Elsevier)Under review · contributor

Reproducible Benchmarking and Model Selection for Art-Style Recognition Across CNN–ViT–Hybrid Backbones

S. Samin, A. F. H. Dhrubo, et al.

Targeting CVPR 2027

Unified Foundation Model for Tracking, Detection, and Segmentation

A single architecture trained jointly across more than 15 large-scale datasets — including Objects365, COCO, and Cityscapes — for multi-class perception under shared representations.

In progressCivil Justice AI

Kolmogorov–Arnold Networks for LLM Frameworks

Re-engineering KAN blocks for transformer language models to improve efficiency and reduce algorithmic bias in Bangladesh's civil judiciary — extending the legal-document classification line from my undergraduate thesis (Themis) toward ACL-grade evaluation.

Edge AIEmbodied

Edge-Native Autonomous Manipulation on Raspberry Pi 5

A framework combining imitation learning, reinforcement learning, and vision-language grounding to enable high-level robotic intelligence without cloud inference — built around the Raspberry Pi 5 as a deployment target for resource-constrained settings.