
Md. Momenul Haque
Lecturer
At a glance
Biography
Driven by a passion for teaching and research, I bring refined analytical skills honed through competitive programming and a deep commitment to academic excellence. As an aspiring lecturer, I aim to create a dynamic learning environment where theoretical knowledge intersects with real-world applications. My research interests lie in cutting-edge fields such as OCR, Blockchain, ML, AI, Computer Vision, NLP, and Medical Image Processing, where I strive to push the boundaries of innovation. I am dedicated to inspiring students by combining practical insights with a love for continuous learning, equipping them with the skills to excel in an evolving technological landscape.
Education
| Degree | Institution | Year | Result |
|---|---|---|---|
| M.Sc.Engg. in Computer Science and Engineering | Rajshahi University of Engineering & Technology, Rajshahi-6204, Bangladesh | 2026 | 3.75/4.00 |
| B.Sc. in Computer Science and Engineering | Bangladesh Army University of Engineering & Technology, Natore-6431, Bangladesh | 2022 | 3.92/4.00 |
| HSC | New Govt. Degree College, Rajshahi, Bangladesh | 2016 | 5.00/5.00 |
| SSC | Seroil Govt. High School, Rajshahi, Bangladesh | 2014 | 5.00/5.00 |
Research Interests
Publications
Journal Publications
Hand gestures and sign language are crucial modes of communication for deaf individuals. Since most people can't understand sign language, it's hard for a mute and an average person to talk to each other. Because of technological progress, computer vision and deep learning can now be used to count. This paper shows two ways to use deep knowledge to recognize sign language. These methods help regular people understand sign language and improve their communication. Based on American sign language (ASL), two separate datasets have been constructed; the first has 26 signs, and the other contains three significant symbols with the crucial sequence of frames or videos for regular communication. This study looks at three different models: the improved ResNet-based convolutional neural network (CNN), the long short-term memory (LSTM), and the gated recurrent unit (GRU). The first dataset is used to fit and assess the CNN model. With the adaptive moment estimation (Adam) optimizer, CNN obtains an accuracy of 89.07%. In contrast, the second dataset is given to LSTM and GRU and a comparison has been conducted. LSTM does better than GRU in all classes. LSTM has a 94.3% accuracy, while GRU only manages 79.3%. Our preliminary models' real-time performance is also highlighted.
Conference Publications
Every day thousands or millions of digital content or assets have been created and transected through the digital marketplace. Most of the cases after creating the content in the digital market a seller completely lose their content ownership and creative content. Not only that the entire system is centralized which means the whole transaction, information sharing, and database update are maintained by the third-party organization. And no contact verification and authentication happen between two parties. Also, generate the possibilities of data manipulation risk for a large amount of data. To solve this problem, we have introduced Ethereum blockchain-based digital marketplace where we use NFT-based smart contracts to verify and authenticate every transaction between buyer, seller, and marketplace owner. Our proposed system operated as a decentralized and distributed ledger. To develop the NFT-based smart contract we use the ERC721 token standard which provides us the non-fungible properties. We use two smart contracts to develop the whole infrastructure. One is for creating a market item and another one is for the marketplace. Also, develop the payment mechanism for our proposed structure and collect the listing fees. So, the proposed Ethereum blockchain-based digital marketplace is secure, trusted, immutable, and properly authenticated by the decentralized network.
Skin cancer, a prevalent and potentially life-threatening disease, necessitates accurate and timely diagnosis for effective treatment. In addressing this critical healthcare challenge, this research proposes a novel approach to skin cancer classification. The study identifies the need for enhanced diagnostic accuracy and introduces a solution by integrating the efficiency of transfer learning, specifically utilizing the EfficientNet V2 architecture. The proposed methodology leverages pre-trained convolutional neural networks (CNNs) to extract high-level features from dermatological images, which are subsequently input into a sophisticated classifier, forming a robust framework for precise skin cancer classification. The research employs a diverse dataset comprising 3297 images sourced from the publicly available Skin Cancer ISIC archive dataset, ensuring comprehensive coverage of benign and malignant skin lesions. During the training phase, the proposed EfficientNet V2 transfer learning approach outperforms conventional methods, achieving an impressive accuracy of 84%, surpassing both Inception V3 (82%) and a generic CNN model (81%). Interpretability analysis is conducted to elucidate the decision-making process of the EfficientNet V2 model, providing insights into the key features influencing its predictions. This transparency is crucial for building trust in the model’s recommendations and facilitating its seamless integration into clinical workflows. The EfficientNet V2 transfer learning approach emerges as a promising and effective solution for advancing skin cancer classification systems, contributing significantly to the ongoing efforts to improve diagnostic accuracy in dermatology.
Falls can cause serious injuries and even death, especially in older adults. Therefore, fall detection systems have the potential to save lives and improve healthcare outcomes. Security and healthcare issues are of utmost concern today in a dynamic world. Accidents from falling can cause serious injury to elderly people or children, and in some cases, death can also occur. A deep learning-based Human Motion Recognition approach has been proposed to detect human falling from video, even in low resolution, and send an alert message to the other end-user without using any costly wearable device. In this paper, the Long-term Recurrent Convolutional Network (LRCN) model is practiced on our proposed FD503 video dataset. A 90% accuracy was attained using the suggested strategy. Precision and recall are 0.93 and 0.85 respectively. The LRCN model is flexible for vision-related problems involving time-varying sequences. The proposed system is highly capable of real-time monitoring to provide healthcare with an alert message.
Leukemia, a malignant blood cancer, is characterized by the uncontrolled proliferation of abnormal white blood cells in the bone marrow and blood. It is a life-threatening disease with various subtypes, making precise classification crucial for effective treatment. However, accurate classification of leukemia subtypes from abnormal cell images poses a challenging problem due to the complex and subtle variations in cell morphology. In response to this challenge, this comprehensive study employs an innovative machine learning approach, utilizing the EfficientNet B3 architecture. With this model, we achieved an impressive accuracy rate of 98.95% and a minimal loss of 0.120, demonstrating its efficacy in classifying leukemia subtypes. To conduct our experiments, we utilized the Kaggle Leukemia cancer dataset, ensuring a robust and reliable evaluation of our proposed methodology. This research contributes significantly to the development of efficient tools for early diagnosis and personalized treatment strategies for leukemia.
Brain tumors represent a critical neurological threat, disrupting brain function due to abnormal tissue growth arising from uncontrolled cell division. Addressing the need for accurate detection, this study proposes an efficient machine-learning approach for brain tumor diagnosis using Magnetic Resonance Imaging (MRI) data. Recent advancements in deep learning, especially Convolutional Neural Networks (CNNs), have shown substantial promise in medical imaging for disease diagnosis. Leveraging CNN s for feature extraction, our research introduces an innovative framework that combines CNN-based analysis with data augmentation and ensemble learning (Random Forest) to classify brain MRI scans as either cancerous or noncancerous with high precision. Our proposed model achieves significant accuracy, with CNN yielding 86% and the integrated CNN-Random Forest model achieving 96%, demonstrating an improvement over traditional pre-trained models with lower computational demands. This approach presents a practical tool for neurosurgeons, supporting rapid, accurate tumor detection and improving diagnostic outcomes in clinical settings. By integrating data augmentation with ensemble learning, our model not only enhances classification performance but also reduces the computational load, making it suitable for real-time applications. This framework thus represents a promising solution to the challenges of brain tumor diagnosis, providing both efficient and reliable detection that may enhance decision-making in medical practice.
South Asia countries like Bangladesh, India, and Pakistan have a large number of fuel filling stations that use centralized payment transaction systems. In some cases, this fuel filling station uses the hand cash payment system which is not secured and time-consuming. Each transaction takes more than five minutes to process. For that reason, in some cases, customers face the huge hassle of standing in a long line and waiting for their turn. Not only that, there are high possibilities of fraud activities and robbery being occur for large amounts of the payment transaction. To solve this problem we propose a blockchain-based payment transaction method for fuel filling stations. Here we use the decentralized open ledger infrastructure and proof-of-work to approve each transaction block. Every transaction between the customer and the filling station authority is completed through a digital wallet which is fully secured, fast, and transparent. Comparing to the bank payment transaction system our proposed method is decentralized and has low transaction fees applied in every transaction. This transaction process is free from third-party involvement and all transactions are immutable. For that reason no issues of customer trust and safe from fraud activities in a large number of payment transactions. Our proposed payment transaction method can play an important part to handle large amounts of transactions and provide transaction security for increasing the number of fuel filling stations in South Asia's most populated country.
Efficient currency detection and counting systems are crucial for the smooth operation of any financial ecosystem. This paper explores the development, implementation, and potential impact of a specialized currency detection and counting system for Bangladeshi coin currencies. Through this comprehensive study, this paper addresses the unique challenges and opportunities in Bangladesh’s coin currency management landscape, highlighting the motivations and results of such a system’s introduction. Computer vision technology includes currency detection. In this paper, we present a coin currency detection system that can detect coin currency from images. Designing a CNN-based coin recognition system for the recognition of Bangladeshi coins of denominations ‘1tk’, ‘2tk’, and ‘5tk’. We have created a dataset that contains more than 10,000 images. We photographed coin currencies from both sides, and the system can recognize coin currencies from both sides. For high efficiency, we test on various backgrounds. Experimental results are presented successfully with a training accuracy of 99.99% with VGG16, 99.38% with VGG19, and 99.99% with EfficientNetB0. We also got validation accuracy with VGG16 of 90.90%, 99.96% with VGG19, and 99.51% with EfficientNetB0.
Human being are able to express their thoughts, perspectives, and opinions on a variety of topics through online forums, blogs, social media sites, digital discussion forums, as well as other virtual spaces as a result of the lightning-fast growth of technological advancements like the Internet and the rise of social networks. These viewpoints may help businesses like hotels and the tourism industry in enhancing their commodities and amenities for the visitors they serve. People are able to learn considerable information from the analysis of public sentiment, thoughts, perceptions, opinions, and so on. Sentiment analysis is essential for exploring the public’s sentiment or opinion because the findings may provide us with insightful information. By using Deep Learning technology to analyze the emotional tenor of comments, it is now possible to quickly understand public opinion on hotel product preferences, advertising tactics, and financial goals. Sentiment Analysis is another Natural Language Processing (NLP) application that has encountered an immense surge in popularity over the past few years. Nevertheless, the difficulties in NLP are impeding sentiment analysis’s efficacy and accuracy. Deep learning models have been highlighted to be a promising response to NLP’s issues lately. This research compares the BERT model with the LSTM model to accomplish sentiment analysis objectives to overcome this limitation. We developed a customized unique data set based on the general consensus regarding hotel reviews to properly implement the aforementioned method. Last but not least, a comparative study between the BERT and LSTM models has been conducted on the experimental results for the input features.
The adoption of charging vehicles in Rajshahi has emerged as a significant step towards reducing air pollution and promoting a healthy, green city. However, the current centralized licensing system poses challenges such as time-consuming processes and the potential for fraudulent activities. To address these issues, this paper proposes the implementation of a secure and decentralized smart licensing system for charging vehicles in Rajshahi City Corporation, leveraging the smart capabilities of blockchain technology. By utilizing Hyperledger Fabric, the implementation generates blockchain based smart contracts that automate and streamline the licensing process smartly and efficiently. This approach ensures the immutability, transparency, and integrity of license records, minimizing the risk of fraud and enhancing overall efficiency. Additionally, the system incorporates the use of QR codes and an inquiry system to enhance the credibility of paper-based licenses. Compared to other digital licenses, this blockchain-based method significantly reduces risks and contributes to the development of a greener and more sustainable Rajshahi city environment. Furthermore, the introduction of a sustainable business policy, utilizing the G2C model, facilitates revenue generation and distribution fairly and equitably for all stakeholders involved in the charging vehicle ecosystem. This implementation not only offers cost benefits and efficiency improvements for drivers but also promotes a sustainable and environmentally friendly city.
The advent of quantum computing necessitates a paradigm shift in securing Firmware-Over-The-Air (FOTA) updates for Internet of Things (IoT) devices. To address this issue, we propose and evaluate a novel quantum-resistant FOTA architecture that integrates three key technologies: the CRYSTALS-Dilithium-5 signature scheme for post-quantum authentication, leveraging an Ethereum smart contract for immutable firmware registration, and the InterPlanetary File System (IPFS) for resilient distribution. Our prototype, benchmarked over 1,000 iterations in a browser environment, validates the practicality of this approach. We measured an average signature verification time of 0.1214 ms for Dilithium, providing 192-bit post-quantum security while incurring a manageable 1.5 × performance overhead compared to the quantum-vulnerable ECDSA P-256 standard. While publisher-side signing incurs a 7.7 × overhead, yet the sub-millisecond verification confirms client-side feasibility. This paper presents the full system design, a comprehensive performance analysis, and discusses adapting this model to embedded Linux platforms (e.g., Raspberry Pi) and the challenges for resource-constrained microcontrollers.
Bangla, one of the most widely spoken languages globally with approximately 265 million speakers, holds profound cultural and linguistic significance. Bangla handwritten character recognition (BHCR) is a vital step in the development of Optical Character Recognition (OCR) systems for preserving and digitizing this rich linguistic heritage. BHCR has critical applications in fields such as document digitization, educational tools, and automation in government and private sectors. However, BHCR presents unique challenges due to the intricate nature of the script, including visually similar characters, complex compound shapes, and a lack of balanced datasets for effective training. Traditional Convolution Neural Network (CNN) and transfer learning methods often struggle to address these complexities adequately. To overcome these limitations, we introduce SondhiNet, a novel framework that integrates CNN-based feature extraction with dimensionality reduction through Principal Component Analysis (PCA), followed by classification using Decision Trees (DT) and Support Vector Machines (SVM). SondhiNet improves accuracy by leveraging an ensemble (voting) mechanism to aggregate predictions effectively. Evaluated on three balanced datasets—Ekush with 122 classes, BanglaLekha with 84 classes, and CMATERdb with 231 classes—SondhiNet achieved accuracies of 97.92%, 97.64%, and 98.47%, respectively. These results underscore the effectiveness of our approach in addressing the challenges of Bangla handwritten character recognition and advancing its practical applications.
Bangla handwritten digit recognition is crucial for various applications, particularly in the financial sector, such as bank cheque processing, automated form filling, postal code recognition, and document verification, where accurate identification of handwritten digits ensures proper transaction handling. English handwritten digit recognition has made significant progress thanks to well-established datasets and methods. However, Bangla handwritten digit recognition remains challenging due to variations in handwriting styles, limited annotated datasets, and the complexity of the Bangla script. To overcome these challenges, we introduce D3Net, an ensemble voting method built on three distinct lightweight Convolutional Neural Network (CNN) architectures optimized using the Adam optimizer. Trained on the NumtaDB dataset, D3Net achieves a remarkable accuracy of 99.52% on the testing set and a Cohen’s Kappa score of 0.99, demonstrating its effectiveness in recognizing Bangla handwritten digits. The system utilizes a multi-CNN approach where each CNN extracts unique features, and the ensemble method combines their predictions to improve overall performance. This robust approach significantly improves recognition accuracy and reliability by effectively addressing the key challenges of handwriting variability, limited datasets, and the complex nature of Bangla digits, ensuring more consistent and precise results in recognition tasks.
Bangla shorthand handwritten characters have important applications in government and administrative work, including rapid transcription, official documentation, and archival record preservation. However, recognizing these characters is difficult because there are few annotated datasets, low-quality images, and limited variation in handwriting styles, which reduces the performance of traditional recognition systems. To solve these problems, we propose a Siamese Network-based Few-Shot Learning approach for recognizing characters in different N-way, N-shot settings. During training, triplet samples consisting of anchor, positive, and negative images are created to help the model learn the similarity between character classes. A similarity comparison method is then used to classify input images accurately, even with only a few training examples. Experiments were conducted on a publicly available Bangla shorthand handwritten character dataset containing 25 classes and 25,000 samples. The dataset was carefully preprocessed and augmented to improve image quality and increase data diversity. The proposed method achieved excellent performance, with 99.93% testing accuracy and an AUC of 1.00 in the 5-way 1-shot setting. In addition, it outperformed the standard Siamese Network and other N-way, N-shot configurations. Compared with benchmark transfer learning models such as ResNet50, MobileNetV2, VGG16, and InceptionV3, the proposed approach showed better accuracy and robustness, making it a reliable solution for Bangla shorthand handwritten character recognition.
Masked face recognition is vital for public safety, yet challenging due to occluded facial features and variations in pose, lighting, and mask type. In this study, we have conducted two distinct experiments. The first experiment focuses on detecting the existence of a face mask on a subject, referred to as Face Mask Identification (FMI). The second experiment aimed to identify the person while wearing a mask, referred to as Masked Face Recognition (MFR). We propose a lightweight transfer learning approach using EfficientNetB0 on two public datasets MDMFR and FMD. For the first experiment, we have trained and tested three models: CNN, ResNet50, and EfficientNetB0. Among them, EfficientNetB0 achieved 100% accuracy on both datasets. To evaluate the model’s generalization ability across datasets, we have conducted cross-dataset validation by training and validating on one dataset (FMD or MDMFR) and testing on the other, then reversing the roles of the datasets to ensure a comprehensive evaluation of performance. The results showed that EfficientNetB0 yielded the highest average accuracy of 97.00%, outperforming CNN (85.00%) and ResNet50 (94.50%). For the second experiment, conducted on MDMFR dataset, EfficientNetB0 achieved superior accuracy of 96.43% which is benchmarked with other state of the art works as well. Overall, this study demonstrates that due to its low parameter count and reduced training time, EfficientNetB0 is well-suited for deployment in resource-constrained environments.
Drowsy driving is a serious road safety problem which causes accidents, injuries, and deaths every year. To help prevent this, it is important to create real-time systems that can detect driver fatigue without disturbing the driver. This paper reviews eight recent studies that use deep learning to detect drowsiness from driver images or video. The methods use traditional Convolutional Neural Networks (CNNs) for analyzing images, lightweight models which can run faster, and advanced models like LSTM and GCNs that track how a driver’s behavior changes over time. It also looks closely at transformer-based models, like Vision Transformer (ViT) and Swin Transformer, which are the latest developments. The models were tested on different datasets, like the MRL Eye Dataset to more realistic driving datasets like NTHU-DDD and SUST-DDD. Results show that transformer and GCN-based models reach the highest accuracy, sometimes it goes over 99.00% in controlled tests, but their performance drops when used in real driving conditions. This review shows three main challenges for future work, one is making models more reliable, second adapting them to real-world driving, and third reducing their computational cost for easy deployment. Overall, this paper summarizes current progress and points out some challenges, and suggests directions for building practical and effective drowsiness detection systems.
Book Chapters
One of the most life-threatening diseases in the world is a brain tumor. A brain tumor is an unusual development of cells inside the cerebrum or the skull. Brain tumor detection is one of the most crucial and arduous tasks in medical image processing. The traditional way of diagnosing brain tumors is time-consuming and prone to human error. From this concern, this chapter presents a deep learning-based approach using MRI images to solve this crucial problem in health care. We consider the public brain tumor MRI image dataset. But, the size of this publicly available dataset is very small to improve the model’s performance and detection. We propose a data augmentation technique called the Deep Convolutional Generative Adversarial Network (DCGAN) algorithm that is used to generate new, previously unseen data that is similar to a given training dataset. Then the collected data would be pre-processed by resizing images, normalizing intensities, and removing noise. Lastly, we use a transformer-based model called the Vision Transformer (ViT) model for brain tumor detection. Experimental results show that the proposed ViT with the data augmentation DCGAN algorithm achieves 99.33% accuracy; without the DCGAN data augmentation algorithm, it achieves 86.27% accuracy. According to the analysis, the proposed ViT model has better accuracy with a minimum training loss of 0.0610, which is small compared to other transfer learning approaches and another deep learning approach. The DCGAN-based training data improved the ViT model’s performance compared to the previous studies. So, the proposed hybrid approach enhances the performance of automatic detection of brain tumors in medical image processing for next-generation medical health care.
The COVID-19 pandemic has triggered an abrupt transformation in the way education is delivered, leading to an unexpected transition to distance learning and creating unique challenges in comprehending student engagement and performance. By accurately classifying student brainwave recordings during class lectures, the challenge of understanding and enhancing student participation can be effectively addressed. We propose a solution based on transfer learning and a principle feature classifier using scalogram images from electroencephalogram (EEG) signals to address the understanding. We use an EEG signal distance learning dataset and preprocess it by removing powerline noise using a bandpass filter, removing artifacts using independent component analysis (ICA), and converting the 1D signal to 2D scalogram EEG images using continuous wavelet transform (CWT). We trained the scalogram images using ResNet-50 and MobileNet V1 transfer learning architectures, outperforming other approaches by leveraging pre-existing knowledge. To extract the primary features, we employ principal component analysis (PCA), a technique that reduces the dimensionality of the data while retaining the crucial features. To determine the level of student engagement during distance learning, as the primary feature classifiers, we employed three traditional machine learning (ML) approaches: K-nearest neighbors (KNN), logistic regression (LR), and support vector machine (SVM). Our experimental result proves that, among these algorithms, ResNet-LR achieved the highest accuracy of https://www.w3.org/1998/Math/MathML" display="inline"> 98 % . The results demonstrate the evaluation of our proposed solution in accurately classifying EEG signals during distance learning, which can aid in understanding student engagement and performance.
Academic Experience
Lecturer
7th May, 2023 to 22th June, 2024Department of CSE Rabindra Maitree University, Kushtia-7000 Duration: 1 year 1 month
Honors & Awards
DR. FATEMA RASHID BEST PAPER AWARD 2024
ain tumor detection on MRI images using a combination of CNN and ensemble learning approach Venue: RU Campus
Dean’s Award- Bangladesh Army University of Engineering & Technology (BAUET)- 2022
Dean’s Award- Bangladesh Army University of Engineering & Technology (BAUET)- 2022
Finalist- Blockchain Olympiad Bangladesh-2024, 2023, 2022
Finalist- Blockchain Olympiad Bangladesh- 2024, 2023, 2022
Honorable Mention- International Collegiate Programming Contest (ICPC)-2026(Coach, BAUET), 2023 (Coach, RMU), 2022
Honorable Mention- International Collegiate Programming Contest (ICPC)-2026(Coach, BAUET), 2023 (Coach, RMU), 2022
Participant-12th National Undergraduate Mathematics Olympiad 2021
Portfolio Startup- Startup Rajshahi- 2022
Portfolio Startup- Startup Rajshahi- 2022
Global Champion (Best Mission Concept)- NASA Space Apps Challenge- 2021
Global Champion (Best Mission Concept)- NASA Space Apps Challenge- 2021
Top 6 (Regional)- Hult Prize Impact Summit HO CHI MINH City (Vietnam)- 2021
Top 6 (Regional)- Hult Prize Impact Summit HO CHI MINH City (Vietnam)- 2021
Branch Ambassador- IEEE Xtreme 15.0 Programming- 2021
Branch Ambassador- IEEE Xtreme 15.0 Programming- 2021
Participant- National Collegiate Programming Contest (NCPC)- 2020
Participant- National Collegiate Programming Contest (NCPC)- 2020
Training
Workshop on Method of Instructions (MOI)
Teaching-Learning in OBE
Quality Assurance in the OBE Framework
Contact
Send a message to Md. Momenul Haque through the department office. Direct email and phone details are listed above.