Rashik Rahman
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Rashik Rahman has a total of 66 co-authors. Here are the top co-authors by number of shared publications:
- Rajibul Islam
Department of Computer Science and Engineering, Bangladesh University of Business and Technology, Dhaka, Bangladesh
5 shared publications - Tanjina Helaly
Department of Computer Science and Engineering, University of Asia Pacific, Dhaka, Bangladesh
5 shared publications - Sk. Tanzir Mehedi
Queensland University of Technology, Brisbane, QLD, Australia
3 shared publications - Ziaur Rahman
Department of Information and Communication Technology, Mawlana Bhashani Science and Technology University, Tangail 1902, Bangladesh
3 shared publications - Abdul Fattah Rakib
Department of Computer Science and Engineering, University of Asia Pacific, Dhaka, Bangladesh
2 shared publications - Ashraful Haque
University of Asia Pacific
2 shared publications
Latest Publications
- 1 December 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Journal of Biomedical and Health Informatics
- Vol. 29 (12) , 8823-8830
Federated Pseudo-Labeling: A Data-Centric, Privacy-Preserving Framework for Medical Image Segmentation
Although essential in the medical domain, protecting patient privacy often restricts data sharing across institutions. Moreover, publicly available datasets often suffer from poor and inconsistent annotation-particularly for image segmentation that requires precise pixel- or voxel-level annotations. As a result, deep learning models are frequently trained on single-institution datasets that are small and lack the heterogeneity of broader patient populations, which limits their generalizability. Federated learning (FL) enables collaborative model training across institutions by sharing model parameters instead of raw medical data. However, it requires uniform model architectures, which may not align with local hardware or software, and still exposes privacy risks through parameter sharing. Further, coordination across institutions with varying data volumes and annotation standards remains challenging, and exchanging model weights-especially for large models-is costly and slow. To address these limitations, we propose DCFed, a data-centric, semi-supervised framework that avoids sharing private data and model parameters by leveraging pseudo-labeling and uncertainty estimation on publicly available unannotated datasets. In our experiments, we use a modified U-Net with residual blocks, atrous spatial pyramid pooling, and convolutional block attention modules at the client level. DCFed improves performance by up to 8.9% on a breast cancer ultrasound dataset and 3.7% on a skin cancer dermoscopy dataset over local training. Notably, DCFed outperforms conventional FL methods such as FedAvg and FedNova across multiple clients in both tasks. In conclusion, DCFed surpasses both centralized training on local datasets and parameter-sharing FL approaches across institutions, establishing a scalable and privacy-preserving solution for real-world medical image segmentation.
- 1 December 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Journal of Biomedical and Health Informatics
- Vol. 29 (12) , 8823-8830
Federated Pseudo-Labeling: A Data-Centric, Privacy-Preserving Framework for Medical Image Segmentation
Although essential in the medical domain, protecting patient privacy often restricts data sharing across institutions. Moreover, publicly available datasets often suffer from poor and inconsistent annotation-particularly for image segmentation that requires precise pixel- or voxel-level annotations. As a result, deep learning models are frequently trained on single-institution datasets that are small and lack the heterogeneity of broader patient populations, which limits their generalizability. Federated learning (FL) enables collaborative model training across institutions by sharing model parameters instead of raw medical data. However, it requires uniform model architectures, which may not align with local hardware or software, and still exposes privacy risks through parameter sharing. Further, coordination across institutions with varying data volumes and annotation standards remains challenging, and exchanging model weights-especially for large models-is costly and slow. To address these limitations, we propose DCFed, a data-centric, semi-supervised framework that avoids sharing private data and model parameters by leveraging pseudo-labeling and uncertainty estimation on publicly available unannotated datasets. In our experiments, we use a modified U-Net with residual blocks, atrous spatial pyramid pooling, and convolutional block attention modules at the client level. DCFed improves performance by up to 8.9% on a breast cancer ultrasound dataset and 3.7% on a skin cancer dermoscopy dataset over local training. Notably, DCFed outperforms conventional FL methods such as FedAvg and FedNova across multiple clients in both tasks. In conclusion, DCFed surpasses both centralized training on local datasets and parameter-sharing FL approaches across institutions, establishing a scalable and privacy-preserving solution for real-world medical image segmentation.
- Open Access
- 10 July 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Data Descriptions
- Vol. 2, 254-261
Descriptor: Bengali Sign Word Recognition Video Dataset (BdSLW102)
Effective communication is the key component of human connection. However, individuals within the deaf and mute community face significant challenges due to the limited accessibility of their unique sign language. To address this issue, sign language emerges as an invaluable asset. Sign language recognition technology has the potential to bridge this communication gap. We have developed an extensive Bengali Word-Level Sign Video Dataset to address this crucial issue. This initiative has the potential to revolutionize our understanding of Bengali sign language. This dataset comprises two parts: words and sentences, each of which includes two sections, the raw video dataset and the masking dataset. The Masking Video Dataset has two types: with background and without background. For the word part, each section contains 102 classes with approximately 300 videos per class, resulting in approximately 91,800 videos overall. For the sentence part, each section comprises 20 classes with approximately 100 videos per class, resulting in approximately 6000 videos overall. The initial phase involved the selection of words and sentences to ensure the relevance of the dataset for real-life communication. A group of volunteers recorded hand gesture signs using smartphones and webcams. We processed the collected dataset using MediaPipe to convert the raw data into masked representations. Participant anonymity is ensured by using Haar Cascade detection, Gaussian blur, and AES encryption. Face data are securely encrypted and can be reconstructed by authorized users. This dataset is freely available for researchers, with dedicated models for recognition and identification. Our work holds the potential to transform communication for the deaf and mute community, advancing sign language recognition and improving lives.
- 10 July 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Data Descriptions
- Vol. 2, 254-261
Open AccessDescriptor: Bengali Sign Word Recognition Video Dataset (BdSLW102)
Effective communication is the key component of human connection. However, individuals within the deaf and mute community face significant challenges due to the limited accessibility of their unique sign language. To address this issue, sign language emerges as an invaluable asset. Sign language recognition technology has the potential to bridge this communication gap. We have developed an extensive Bengali Word-Level Sign Video Dataset to address this crucial issue. This initiative has the potential to revolutionize our understanding of Bengali sign language. This dataset comprises two parts: words and sentences, each of which includes two sections, the raw video dataset and the masking dataset. The Masking Video Dataset has two types: with background and without background. For the word part, each section contains 102 classes with approximately 300 videos per class, resulting in approximately 91,800 videos overall. For the sentence part, each section comprises 20 classes with approximately 100 videos per class, resulting in approximately 6000 videos overall. The initial phase involved the selection of words and sentences to ensure the relevance of the dataset for real-life communication. A group of volunteers recorded hand gesture signs using smartphones and webcams. We processed the collected dataset using MediaPipe to convert the raw data into masked representations. Participant anonymity is ensured by using Haar Cascade detection, Gaussian blur, and AES encryption. Face data are securely encrypted and can be reconstructed by authorized users. This dataset is freely available for researchers, with dedicated models for recognition and identification. Our work holds the potential to transform communication for the deaf and mute community, advancing sign language recognition and improving lives.
- Open Access
- 4 March 2025
- Published by Elsevier in Data in Brief
- Vol. 59, 111440
FallVision: A benchmark video dataset for fall detection
- 4 March 2025
- Published by Elsevier in Data in Brief
- Vol. 59, 111440
Open AccessFallVision: A benchmark video dataset for fall detection
- Open Access
- 28 September 2024
- Published by MDPI AG in Mathematics
- Vol. 12 (19) , 3042
An Explainable AI-Based Modified YOLOv8 Model for Efficient Fire Detection
Early fire detection is the key to saving lives and limiting property damage. Advanced technology can detect fires in high-risk zones with minimal human presence before they escalate beyond control. This study focuses on providing a more advanced model structure based on the YOLOv8 architecture to enhance early recognition of fire. Although YOLOv8 is excellent at real-time object detection, it can still be better adjusted to the nuances of fire detection. We achieved this advancement by incorporating an additional context-to-flow layer, enabling the YOLOv8 model to more effectively capture both local and global contextual information. The context-to-flow layer enhances the model’s ability to recognize complex patterns like smoke and flames, leading to more effective feature extraction. This extra layer helps the model better detect fires and smoke by improving its ability to focus on fine-grained details and minor variation, which is crucial in challenging environments with low visibility, dynamic fire behavior, and complex backgrounds. Our proposed model achieved a 2.9% greater precision rate, 4.7% more recall rate, and 4% more F1-score in comparison to the YOLOv8 default model. This study discovered that the architecture modification increases information flow and improves fire detection at all fire sizes, from tiny sparks to massive flames. We also included explainable AI strategies to explain the model’s decision-making, thus adding more transparency and improving trust in its predictions. Ultimately, this enhanced system demonstrates remarkable efficacy and accuracy, which allows additional improvements in autonomous fire detection systems.
- 28 September 2024
- Published by MDPI AG in Mathematics
- Vol. 12 (19) , 3042
Open AccessAn Explainable AI-Based Modified YOLOv8 Model for Efficient Fire Detection
Early fire detection is the key to saving lives and limiting property damage. Advanced technology can detect fires in high-risk zones with minimal human presence before they escalate beyond control. This study focuses on providing a more advanced model structure based on the YOLOv8 architecture to enhance early recognition of fire. Although YOLOv8 is excellent at real-time object detection, it can still be better adjusted to the nuances of fire detection. We achieved this advancement by incorporating an additional context-to-flow layer, enabling the YOLOv8 model to more effectively capture both local and global contextual information. The context-to-flow layer enhances the model’s ability to recognize complex patterns like smoke and flames, leading to more effective feature extraction. This extra layer helps the model better detect fires and smoke by improving its ability to focus on fine-grained details and minor variation, which is crucial in challenging environments with low visibility, dynamic fire behavior, and complex backgrounds. Our proposed model achieved a 2.9% greater precision rate, 4.7% more recall rate, and 4% more F1-score in comparison to the YOLOv8 default model. This study discovered that the architecture modification increases information flow and improves fire detection at all fire sizes, from tiny sparks to massive flames. We also included explainable AI strategies to explain the model’s decision-making, thus adding more transparency and improving trust in its predictions. Ultimately, this enhanced system demonstrates remarkable efficacy and accuracy, which allows additional improvements in autonomous fire detection systems.
- Open Access
- 16 September 2024
- Published by MDPI AG in Mathematics
- Vol. 12 (18) , 2886
Enhancing Autism Spectrum Disorder Classification with Lightweight Quantized CNNs and Federated Learning on ABIDE-1 Dataset
Autism spectrum disorder (ASD) is a complex neurodevelopmental condition that presents significant diagnostic challenges due to its varied symptoms and nature. This study aims to improve ASD classification using advanced deep learning techniques applied to neuroimaging data. We developed an automated system leveraging the ABIDE-1 dataset and a novel lightweight quantized one-dimensional (1D) Convolutional Neural Network (Q-CNN) model to analyze fMRI data. Our approach employs the NIAK pipeline with multiple brain atlases and filtering methods. Initially, the Regions of Interest (ROIs) are converted into feature vectors using tangent space embedding to feed into the Q-CNN model. The proposed 1D-CNN is quantized through Quantize Aware Training (QAT). As the quantization method, int8 quantization is utilized, which makes it both robust and lightweight. We propose a federated learning (FL) framework to ensure data privacy, which allows decentralized training across different data centers without compromising local data security. Our findings indicate that the CC200 brain atlas, within the NIAK pipeline’s filt-global filtering methods, provides the best results for ASD classification. Notably, the ASD classification outcomes have achieved a significant test accuracy of 98% using the CC200 and filt-global filtering techniques. To the best of our knowledge, this performance surpasses previous studies in the field, highlighting a notable enhancement in ASD detection from fMRI data. Furthermore, the FL-based Q-CNN model demonstrated robust performance and high efficiency on a Raspberry Pi 4, underscoring its potential for real-world applications. We exhibit the efficacy of the Q-CNN model by comparing its inference time, power consumption, and storage requirements with those of the 1D-CNN, quantized CNN, and the proposed int8 Q-CNN models. This research has made several key contributions, including the development of a lightweight int8 Q-CNN model, the application of FL for data privacy, and the evaluation of the proposed model in real-world settings. By identifying optimal brain atlases and filtering methods, this study provides valuable insights for future research in the field of neurodevelopmental disorders.
- 16 September 2024
- Published by MDPI AG in Mathematics
- Vol. 12 (18) , 2886
Open AccessEnhancing Autism Spectrum Disorder Classification with Lightweight Quantized CNNs and Federated Learning on ABIDE-1 Dataset
Autism spectrum disorder (ASD) is a complex neurodevelopmental condition that presents significant diagnostic challenges due to its varied symptoms and nature. This study aims to improve ASD classification using advanced deep learning techniques applied to neuroimaging data. We developed an automated system leveraging the ABIDE-1 dataset and a novel lightweight quantized one-dimensional (1D) Convolutional Neural Network (Q-CNN) model to analyze fMRI data. Our approach employs the NIAK pipeline with multiple brain atlases and filtering methods. Initially, the Regions of Interest (ROIs) are converted into feature vectors using tangent space embedding to feed into the Q-CNN model. The proposed 1D-CNN is quantized through Quantize Aware Training (QAT). As the quantization method, int8 quantization is utilized, which makes it both robust and lightweight. We propose a federated learning (FL) framework to ensure data privacy, which allows decentralized training across different data centers without compromising local data security. Our findings indicate that the CC200 brain atlas, within the NIAK pipeline’s filt-global filtering methods, provides the best results for ASD classification. Notably, the ASD classification outcomes have achieved a significant test accuracy of 98% using the CC200 and filt-global filtering techniques. To the best of our knowledge, this performance surpasses previous studies in the field, highlighting a notable enhancement in ASD detection from fMRI data. Furthermore, the FL-based Q-CNN model demonstrated robust performance and high efficiency on a Raspberry Pi 4, underscoring its potential for real-world applications. We exhibit the efficacy of the Q-CNN model by comparing its inference time, power consumption, and storage requirements with those of the 1D-CNN, quantized CNN, and the proposed int8 Q-CNN models. This research has made several key contributions, including the development of a lightweight int8 Q-CNN model, the application of FL for data privacy, and the evaluation of the proposed model in real-world settings. By identifying optimal brain atlases and filtering methods, this study provides valuable insights for future research in the field of neurodevelopmental disorders.
- Open Access
- 4 September 2024
- Published by MDPI AG in Preprints.org
An Explainable AI-based Modified YOLOv8 Model for Efficient Fire Detection
Fire detection is crucial for ensuring human safety and minimizing property damage. By utilizing advanced technologies, we can identify fires early, before they escalate. Autonomous fire detection systems are particularly vital in high-risk areas with minimal human presence. To address this need, we propose an automated fire-detection model based on YOLOv8. We modified the architecture of YOLOv8 and achieved impressive performances. Through comprehensive analysis of different YOLO models and their predecessors, we identified and addressed substantial gaps in this field. Also, by augmenting the framework with additional capabilities, we surpassed previous models, enabling real-time fire detection. Our proposed model achieves impressive performances, with 98% accuracy in fire detection and 97.8% in smoke detection. Moreover, recall rates for fire and smoke are 97.1% and 97.4%, respectively, with a mean average precision (mAP) accuracy of 99.1% for both cases. Finally, we applied an explainable artificial intelligence technique to interpret our proposed model and its results. This study lays the foundation for future research aimed at enhancing fire detection efficiency. The reliability of our modified model presents promising opportunities for future advancements, ensuring increased effectiveness and precision in efficient fire detection.
- 4 September 2024
- Published by MDPI AG in Preprints.org
Open AccessAn Explainable AI-based Modified YOLOv8 Model for Efficient Fire Detection
Fire detection is crucial for ensuring human safety and minimizing property damage. By utilizing advanced technologies, we can identify fires early, before they escalate. Autonomous fire detection systems are particularly vital in high-risk areas with minimal human presence. To address this need, we propose an automated fire-detection model based on YOLOv8. We modified the architecture of YOLOv8 and achieved impressive performances. Through comprehensive analysis of different YOLO models and their predecessors, we identified and addressed substantial gaps in this field. Also, by augmenting the framework with additional capabilities, we surpassed previous models, enabling real-time fire detection. Our proposed model achieves impressive performances, with 98% accuracy in fire detection and 97.8% in smoke detection. Moreover, recall rates for fire and smoke are 97.1% and 97.4%, respectively, with a mean average precision (mAP) accuracy of 99.1% for both cases. Finally, we applied an explainable artificial intelligence technique to interpret our proposed model and its results. This study lays the foundation for future research aimed at enhancing fire detection efficiency. The reliability of our modified model presents promising opportunities for future advancements, ensuring increased effectiveness and precision in efficient fire detection.
- Open Access
- 25 August 2024
- Published by Elsevier in Data in Brief
- Vol. 57, 110875
VID: A comprehensive dataset for violence detection in various contexts
- 25 August 2024
- Published by Elsevier in Data in Brief
- Vol. 57, 110875
Open AccessVID: A comprehensive dataset for violence detection in various contexts
- Open Access
- 20 March 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Open Journal of the Computer Society
- Vol. 5, 120-130
A Lightweight Visual Font Style Recognition With Quantized Convolutional Autoencoder
Font style recognition plays a vital role in the field of computer vision, particularly in document and pattern analysis, and image processing. In the industry context, this recognition of font styles holds immense importance for professionals such as graphic designers, front-end developers, and UI-UX developers. In recent times, font style recognition using Computer Vision has made significant progress, especially in English. Very few works have been done for other languages as well. However, the existing models are computationally costly, time-consuming, and not diversified. In this work, we propose a state-of-the-art model to recognize Bangla fonts from images using a quantized Convolutional Autoencoder (Q-CAE) approach. The compressed model takes around 58 KB of memory only which makes it suitable for not only high-end but also low-end computational edge devices. We have also created a synthetic data set consisting of 10 distinct Bangla font styles and a total of 60,000 images for conducting this study as no dedicated dataset is available publicly. Experimental outcomes demonstrate that the proposed method can perform better than existing methods, gaining an overall accuracy of 99.95% without quantization and 99.85% after quantization.
- 20 March 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Open Journal of the Computer Society
- Vol. 5, 120-130
Open AccessA Lightweight Visual Font Style Recognition With Quantized Convolutional Autoencoder
Font style recognition plays a vital role in the field of computer vision, particularly in document and pattern analysis, and image processing. In the industry context, this recognition of font styles holds immense importance for professionals such as graphic designers, front-end developers, and UI-UX developers. In recent times, font style recognition using Computer Vision has made significant progress, especially in English. Very few works have been done for other languages as well. However, the existing models are computationally costly, time-consuming, and not diversified. In this work, we propose a state-of-the-art model to recognize Bangla fonts from images using a quantized Convolutional Autoencoder (Q-CAE) approach. The compressed model takes around 58 KB of memory only which makes it suitable for not only high-end but also low-end computational edge devices. We have also created a synthetic data set consisting of 10 distinct Bangla font styles and a total of 60,000 images for conducting this study as no dedicated dataset is available publicly. Experimental outcomes demonstrate that the proposed method can perform better than existing methods, gaining an overall accuracy of 99.95% without quantization and 99.85% after quantization.
- 20 September 2023
- Published by Springer Nature in Arabian Journal for Science and Engineering
- Vol. 49 (3) , 4097-4108
A Lightweight Quantized CNN Model for Plant Disease Recognition
- 20 September 2023
- Published by Springer Nature in Arabian Journal for Science and Engineering
- Vol. 49 (3) , 4097-4108
A Lightweight Quantized CNN Model for Plant Disease Recognition
- Open Access
- 18 June 2023
- Published by Elsevier in Data in Brief
- Vol. 49, 109329
BDSL 49: A comprehensive dataset of Bangla sign language
- 18 June 2023
- Published by Elsevier in Data in Brief
- Vol. 49, 109329
Open AccessBDSL 49: A comprehensive dataset of Bangla sign language