Publications

@inproceedings{dab2023novel,
  title={A Novel Deep Learning Approach for COVID-19 Detection from Chest X-rays},
  author={Dab, Rajib and Rahman, Ahsan and Akter, Saima},
  booktitle={International Conference on Artificial Intelligence and Soft Computing},
  pages={123--130},
  year={2023},
  organization={IEEE}
}

The COVID-19 pandemic has created an urgent need for fast and reliable diagnostic tools. In this paper, we propose a novel deep learning approach for detecting COVID-19 from chest X-ray images. Our method utilizes a modified ResNet architecture with attention mechanisms to focus on relevant regions of X-ray images. We evaluate our approach on a dataset of 15,000 X-ray images, achieving 98.5% accuracy, 97.8% sensitivity, and 98.9% specificity. Our method outperforms previous approaches in both accuracy and computational efficiency, making it suitable for real-time applications in resource-constrained healthcare settings. We also provide visualizations of the model's decision process using Grad-CAM, enhancing explainability for healthcare professionals.

@inproceedings{dab2022improved,
  title={Improved Recognition of Handwritten Bangla Text Using Hybrid CNN-RNN Architecture},
  author={Dab, Rajib and Islam, Mohammad},
  booktitle={International Conference on Bangla Speech and Language Processing},
  pages={45--52},
  year={2022},
  organization={ACM}
}

Handwritten text recognition for non-Latin scripts like Bangla presents unique challenges due to the script's complex structure and character variations. This paper presents a hybrid CNN-RNN architecture for recognizing handwritten Bangla text. Our approach combines the spatial feature extraction capabilities of CNNs with the sequential learning abilities of RNNs, specifically using a ResNet backbone followed by Bidirectional LSTM layers. We evaluate our model on the BanglaWriting dataset, containing over 10,000 handwritten Bangla text samples. Our approach achieves a character recognition accuracy of 94.3% and a word recognition accuracy of 87.5%, outperforming previous state-of-the-art methods by 3.2% and 4.1% respectively. We also analyze the impact of various data augmentation techniques and model configurations on recognition performance, providing insights for future research in this area.

@inproceedings{dab2021smart,
  title={Smart Agriculture Monitoring System: An IoT-Based Approach for Sustainable Farming in Bangladesh},
  author={Dab, Rajib and Khanam, Fahmida and Karim, Abdul},
  booktitle={Advances in Artificial Intelligence and IoT Technologies},
  pages={210--225},
  year={2021},
  publisher={Springer}
}

Agriculture is the backbone of Bangladesh's economy, employing nearly half of the country's workforce. However, farmers face significant challenges, including climate change impacts, water scarcity, and limited access to technology. This paper presents an IoT-based smart agriculture monitoring system designed specifically for rural Bangladeshi farmers. The system uses low-cost sensors to monitor soil moisture, temperature, and humidity, providing actionable insights to farmers via a user-friendly mobile application. Our solution incorporates solar-powered nodes with long-range communication capabilities, addressing the unreliable electricity and internet connectivity in rural areas. We evaluate the system through a 6-month field trial involving 50 farmers in the Sylhet region, demonstrating water usage reduction of 30% and yield improvements of 22% compared to traditional farming methods. Cost-benefit analysis shows that farmers can recover the system's cost within one growing season through resource savings and increased productivity. The paper also discusses implementation challenges and provides a roadmap for scaling the solution across Bangladesh's agricultural regions.