2025
- Prince, A.T.Z., Paul, D., Jim, M.I., Haider, I., Goswamee, G., Tuhin, A.H., Ahmed, T., Khoka, Z.H., Hasan, M.M., Hossain, M.L. (2025), A Deep Learning Framework for Reliable Vehicle Identification in Intelligent Transportation Systems. In: Proceedings of the 4th International Conference on Innovations in Data Analytics (ICIDA 2025). see credentials
Research Problem
Reliable vehicle identification is a core challenge in intelligent transportation systems due to occlusion, lighting variation, camera perspective changes, and heterogeneous vehicle types. Existing vision-based methods often fail under real-world traffic conditions or require expensive sensor infrastructure.
What We Did
We designed a custom convolutional neural network (CNN) for binary vehicle/non-vehicle classification and evaluated its performance against widely used deep learning architectures.
What We Did
We designed a custom convolutional neural network (CNN) for binary vehicle/non-vehicle classification and evaluated its performance against widely used deep learning architectures.
- Built and curated a dataset of 3,026 labeled vehicle and non-vehicle images
- Developed a multi-layer CNN using Keras with preprocessing, normalization, and data augmentation
- Conducted a comparative study with InceptionV3 and AlexNet
- Evaluated models using Accuracy, Precision, Recall, and F1-Score
Key Results
- Achieved 97.03% classification accuracy, outperforming InceptionV3 and AlexNet (~92%)
- Demonstrated improved robustness under real-world traffic conditions
- Reduced false positives and improved reliability of vehicle identification
My Contribution
I contributed to CNN model development, experimental evaluation, and comparative performance analysis, and participated in preparing and presenting the work at ICIDA 2025.
- Tuhin, A.H.(2025), Optimized Transport Layers for Enhanced Efficiency in Perovskite-Perovskite Tandem Solar Cells. See credentials
Research Problem
Perovskite–perovskite tandem solar cells have the theoretical potential to exceed the efficiency limits of conventional silicon solar cells, but their practical performance is often constrained by optical losses, inefficient charge transport, and current mismatch between sub-cells. Optimizing electron and hole transport layers is critical to unlocking higher power conversion efficiency (PCE).
What We Did
This work investigates how optimized electron transport layers (ETL) and hole transport layers (HTL) can enhance photon coupling, charge extraction, and overall efficiency in 2-terminal perovskite–perovskite tandem solar cells.
- Designed a 2-terminal tandem device architecture with wide-bandgap and low-bandgap perovskite absorbers
- Optimized AZO-based ETL and NiO/AZO-based HTL to reduce recombination and optical losses
- Tuned absorber bandgaps (~1.72 eV top cell, ~1.17 eV bottom cell) for effective spectrum utilization
- Evaluated quantum efficiency, refractive indices, and power density distribution using optical and device simulations
Key Results
- Achieved a power conversion efficiency of ~22.2%
- Obtained a combined open-circuit voltage (Voc ≈ 1.85 V) through effective tandem integration
- Demonstrated improved current matching, photon harvesting, and reduced optical loss
- Showed that optimized transport layers significantly enhance device performance compared to existing tandem designs
My Contribution
I contributed to the conceptual design of the tandem architecture, transport-layer optimization strategy, and analysis of optical and electrical performance, and participated in interpreting simulation results and preparing the manuscript.
- Tuhin, A.H., Barua, S. (2025). Predictive Modeling of Renewable Energy Generation to Reduce CO₂ Emissions for Net Zero in Bangladesh Using Hybrid Models. Poster presented at The 1st International Online Conference on Designs, MDPI. See credentials
Research Problem
Achieving net-zero emissions under the Paris Agreement requires accurate long-term forecasting of both CO₂ emissions and the renewable energy capacity needed to offset them. In Bangladesh, national-level forecasting studies remain limited, and traditional statistical models struggle to capture the nonlinear, uncertain dynamics of energy–emission systems.
What This Research Studies
This work investigates whether hybrid predictive models—combining machine learning, deep learning, and probabilistic approaches—can significantly improve forecasting accuracy and uncertainty estimation for renewable energy generation and CO₂ emissions in Bangladesh.
Approach & Methodology
We developed and evaluated three hybrid forecasting frameworks using long-term national datasets:
- LSTM + ARIMA + XGBoost to jointly model linear trends, nonlinear patterns, and temporal dependencies
- DNN + Gaussian Process Regression (GPR) to capture complex relationships with probabilistic uncertainty estimation
- Random Forest + LSTM + Bayesian Neural Network (BNN) for ensemble-based robustness
The models were trained on 53 years of national renewable energy generation data (1971–2023) and CO₂ emissions data (1950–2023) sourced from Our World in Data. Data preprocessing, exploratory analysis, and train–test splitting (80/20) were applied consistently across models.
Key Findings (Preliminary)
- The DNN–GPR hybrid model achieved the highest predictive performance, with near-zero error and strong uncertainty calibration
- Ensemble hybrid models (LSTM–ARIMA–XGBoost and RF–LSTM–BNN) provided more conservative forecasts, useful for short-term volatility and structural-break detection
- Results suggest hybrid and probabilistic models substantially outperform conventional forecasting approaches in the Bangladesh context
My Contribution
I am the first author and led the formulation of the problem, design of the model, data analysis, and evaluation, focusing on linking forecasting outputs to policy-relevant net-zero planning for Bangladesh’s energy system.
Under Review
- Tuhin, A.H. (2025). Design and Development of a Soft Robotic Endoscope with Adaptive Stiffness for Gastrointestinal Examination. See credentials
Manuscript In Preparation
- Fattah, A., Mehnaj, I., Iqra, Tuhin, A.H. (2025). An Analysis to Assess Reasons Behind High Tariff Rates of Solar IPP in Bangladesh.
- Tuhin, A.H. (2025). CNN-Based Vehicle Detection and Classification in Complex Traffic Environments: A Comparative Study with InceptionV3.