Computer Science and Engineering

Browse

Now showing 1 - 20 of 50

An Exploratory Approach to Find a Novel Metric Based Optimum Language Model for Automatic Bangla Word Prediction
(International Journal of Intelligent Systems and Applications, 2020-09-27)
Word completion and word prediction are two important phenomena in typing that have intense effect on aiding disable people and students while using keyboard or other similar devices. Such auto completion technique also helps students significantly during learning process through constructing proper keywords during web searching. A lot of works are conducted for English language, but for Bangla, it is still very inadequate as well as the metrics used for performance computation is not rigorous yet. Bangla is one of the mostly spoken languages (3.05% of world population) and ranked as seventh among all the languages in the world. In this paper, word prediction on Bangla sentence by using stochastic, i.e. N-gram based language models are proposed for auto completing a sentence by predicting a set of words rather than a single word, which was done in previous work. A novel approach is proposed in order to find the optimum language model based on performance metric. In addition, for finding out better performance, a large Bangla corpus of different word types is used.
A Proximity Weighted Evidential k Nearest Neighbor Classifier for Imbalanced Data
(2020-10-04) Akash, Pritom Saha; Sharmin, Sadia; Ali, Amin Ahsan; Shoyaib, Mohammad
In k Nearest Neighbor (kNN) classifier, a query instance is classified based on the most frequent class of its nearest neighbors among the training instances. In imbalanced datasets, kNN becomes biased towards the majority instances of the training space. To solve this problem, we propose a method called Proximity weighted Evidential kNN classifier. In this method, each neighbor of a query instance is considered as a piece of evidence from which we calculate the probability of class label given feature values to provide more preference to the minority instances. This is then discounted by the proximity of the neighbor to prioritize the closer instances in the local neighborhood. These evidences are then combined using Dempster-Shafer theory of evidence. A rigorous experiment over 30 benchmark imbalanced datasets shows that our method performs better compared to 12 popular methods. In pairwise comparison of these 12 methods with our method, in the best case, our method wins in 29 datasets, and in the worst case it wins in least 19 datasets. More importantly, according to Friedman test the proposed method ranks higher than all other methods in terms of AUC at 5% level of significance.
Discretization and Feature Selection Based on Bias Corrected Mutual Information Considering High-Order Dependencies
(Springer Link, 2020-10-04) Sharmin, Sadia; Ali, Amin Ahsan; Shoyaib, Mohammad
Mutual Information (MI) based feature selection methods are popular due to their ability to capture the nonlinear relationship among variables. However, existing works rarely address the error (bias) that occurs due to the use of finite samples during the estimation of MI. To the best of our knowledge, none of the existing methods address the bias issue for the high-order interaction term which is essential for better approximation of joint MI. In this paper, we first calculate the amount of bias of this term. Moreover, to select features using χ2 based search, we also show that this term follows χ2 distribution. Based on these two theoretical results, we propose Discretization and feature Selection based on bias corrected Mutual information (DSbM). DSbM is extended by adding simultaneous forward selection and backward elimination (DSbM fb ). We demonstrate the superiority of DSbM over four state-of-the-art methods in terms of accuracy and the number of selected features on twenty benchmark datasets. Experimental results also demonstrate that DSbM outperforms the existing methods in terms of accuracy, Pareto Optimality and Friedman test. We also observe that compared to DSbM, in some dataset DSbM fb selects fewer features and increases accuracy.
Blockchain-Based Intrusion Detection Systems (IDS) using Decentralized Threat Intelligence Sharing
(IUB, 2023-01) Hasan, Mostafa; Alam, Moinul; Akash, Arvil Nath
This paper presents a theoretical analysis of a blockchain-based Intrusion Detection System (IDS) designed to improve cybersecurity through decentralized threat intelligence sharing. The proposed IDS framework integrates the Isolation Forest algorithm, a machine learning model, with blockchain technology to enable anomaly detection and automated threat response across a network. Using the Isolation Forest algorithm, the system effectively identifies anomalies in network traffic, while smart contracts on the blockchain allow for autonomous actions, such as node quarantine and malicious IP blocking. This approach aims to reduce the frequency of false positives and enhance response times, providing an efficient solution for scalable and collaborative cybersecurity applications. By examining the the potential of combining machine learning with blockchain and smart contracts, this paper contributes to the advancement of adaptive IDS solutions that leverage decentralized architectures for enhanced threat detection and rapid response.
An Undergraduate Internship/Project on “Store Point of Sale Solution Using Power App”
(Independent University, Bangladesh, 2023-01-24) Khan, Ahmad Sayeef
Internship is defined as obtaining practical experience from various organizations, which helps in the formation of a connection between theoretical and practical knowledge. It is very important because it is the first time for a student to acquire a keen practical knowledge from the different organizations. When I was offered an internship at TechTrioZ Solutions, I got the chance to work and learn with developer team. The project’s goal was to create a Point of sale system using Power App platform which is very newly used platform in Bangladesh. This report covers the whole project that I learned about throughout my internship period. By the mercy of Allah I had prior development experience and knowledge about the Microsoft Power App, Power Automate, Share-point, Microsoft Data-verse, and Dynamic 365 thus I was able to take responsibility for the development of this application. In first chapter there is an introduction about the project, background of the project, objectives, scope of the project and about the organization where I worked. Chapter two describes the literature review where I discussed about market analysis both in local and global market, about similar products and how my undergraduate studies helps me to do this project. Chapter three describes the project management and financing of the project where I describes work breakdown structure, time distribution show in critical map diagram, gantt chart, activity wise resource allocation and about the budget. Chapter four describes about methodology where I describes about agile methodology which I used here, I also describe why use agile methodology. Chapter five describes body projects, where I describe in details about work description, six element analysis, feasibility analysis, problem, effects and constraints analysis. I also give here rich picture, erd diagram, activity diagram, use case diagram, sequence. Functional, non-functional requirements, input, output and architecture of the project also describes in this section. Chapter 6 describes about survey results and analysis. Chapter seven describes project as engineering problem analysis which includes sustainability of the project, social and environmental effects of the project, addressing ethics and ethical issues. Chapter eight describes about the future work, lesson I learned from my internship and finally the conclusion.
An Undergraduate Internship / Project on “Mobile Application for Comilla University”
(Independent University, Bangladesh, 2023-01-24) PIYAL, AKIB HAMID
Comilla University Application (COU APP) is a mobile application which was built using flutter Node.js, express and MySQL. The application's user interface was made using Flutter. Flutter makes it easier to create cross-platform applications from a single codebase for Android, iOS, Linux, macOS, Windows, and the web. The backend REST API was built using Node.js, Express, and MySQL. The application lets users to stay in touch with university faculty and staff and receive the most recent university updates. Users can download the most recent university announcement. The user can browse all departments and communicate with department faculty members directly from the application by message, phone, or email. Users can view the most recent events taking place at the university. Users can view the holidays and off days on the built-in calendar. Students at Comilla University can view the schedule for the student bus by using the application. They can determine the bus's beginning and terminating points and the times that correspond to them.
An Undergraduate Internship/Project on World Marketing Summit Bangladesh and Bangladesh Institute of Modern Marketing Website
(Independent University, Bangladesh, 2023-01-24) Islam, Md. Towhidul
This report is the outcome of my internship at the World Marketing Summit Bangladesh under the Northern Education Group (NEG), one of the biggest education groups in Bangladesh. I was assigned with the task of updating the content, design, functionalities and layout of the existing website of World Marketing Summit Bangladesh, as well as developing and deploying a new website for “Bangladesh Institute of Modern Marketing (BIMM)” from. For accomplishing both of the tasks assigned to me, I have successfully completed all the necessary steps, starting from requirement gathering, planning, content preparing, designing, implementing and testing. Along with web development, I have also actively participated in other necessary corporate affairs regarding marketing and communications. Working at the Northern Education Group (NEG) for World Marketing Summit has added a new dimension in my career by bestowing me with vast knowledge and experience regarding corporate life and culture. I find myself grateful for this opportunity and wish to carry the knowledge earned through this internship to my future endeavors.
Peering Through the Past: Forensic Analysis of Impression Via Deep Learning
(Independent University, Bangladesh, 2023-08) Shroddha, Sharmin Islam; Shaba, Sumaia Anjum; Titu, Kazi Sohrab Uddin
Indented writing is a crucial part of forensic science because it can reveal hidden informa- tion and evidence that is invisible to the eye. This is achieved by identifying impressions of the original writing that are present beneath the surface of the writing page. Yet, traditional methods for deciphering indented writing, are time-consuming and can pose environmental and health risks. Manual and chemical approaches also carry the risk of evidence distortion. In this study, we propose an innovative approach to deciphering indented writing using semantic segmentation and deep learning. Our objective is to provide a more efficient and non-destructive solution for deciphering indented writing to benefit the forensic domain. To achieve this goal, we carefully created a comprehensive dataset. In our study, we employed the U-Net deep learning model to precisely catego- rize each pixel as either belonging to the indented writing region or not. We assessed our model's performance by employing two distinct image channels: grayscale and color (RGB). When trained with grayscale images, the model had a remarkable accuracy of 98.95% and a Mean Intersection over Union (mIoU) of 70.88%. But when color images were used, the model's performance was greatly improved, with accuracy and mloU of 98.86% and 84.14%, respectively. These findings indicate that accurate indented writing recognition by the model is enhanced by RGB imagery.
Land Use Land Cover Segmentation using Synthetic Aperture Radar Data on Dhaka Division, Bangladesh
(Independent University, Bangladesh, 2023-09) Khan, Sadia; Almas, Mir Sayad Bin
Bangladesh is an agrarian economy where agriculture contributes 19.3% of the gross domestic product (GDP) of the country. [1] Rice is the dominant crop and the staple food that meets the nutritional demand of the large population of the country. It is becoming increasingly important to monitor paddy production in the country and detect threats to the crop production due to climate change, natural disasters, urbanization and other factors in order to utilize governmental aid and resources as efficiently as possible as well as take preventive measures. Clouds cover the sky of Bangladesh from the months of May through October, which overlaps with the major rice crop cultivation season of Aus and Boro. Optical images with optical remote sensing simply fail to capture the condition of the crops during these cloudy months. This research proposes the use of Synthetic Aperture Radar (SAR) images to penetrate through clouds during the cloudy months in Bangladesh. We conducted our research over the area of Dhaka district using Sentinel-1A images collected from Copernicus Open Access Hub website and an annotated Bing image as ground truth with five major areas of interest to create the training and testing datasets to input to a UNET model, which is a popular deep learning architecture for image segmentation tasks. We find out that despite the unfavorable weather and presence of clouds, our model performs really well with a 76% accuracy and promising results to support the usage of Sentinel-1A images to monitor areas during monsoon season.
Music Genre Classification and Sentiment Analysis of Bengali Music based on various inherent audio features
(IUB, CSE, 2024-04-27) Humayra, Atika; Sohag, Md Maruf Kamran
The classification and categorization of Bangla music genres and sentiments play a pivotal role in the development of dedicated online platforms catering to Bengali music enthusiasts. These platforms aim to provide users with a seamless experience, allowing them to explore and discover music that resonates with their tastes and moods. Through effective organization based on genre and sentiment, users can easily navigate vast music libraries, enhancing their overall engagement and satisfaction. While significant strides have been made in the field of music classification and sentiment analysis, there is always room for improvement. Current models have demonstrated promising results, but their performance can be further enhanced through iterative refinement and expansion of the underlying datasets. This continual improvement process is essential for ensuring the accuracy and efficiency of genre classification and sentiment analysis in Bengali music. One approach to improving model performance is the gradual tuning of hyperparameters. Hyperparameters are crucial settings that govern the behavior of machine learning models, and fine-tuning them can have a significant impact on performance. By systematically adjusting these parameters over time, researchers can optimize model performance to achieve better results in genre classification and sentiment analysis tasks. This iterative tuning process allows for the identification of optimal parameter configurations that maximize model accuracy and efficiency. In addition to hyperparameter tuning, expanding the dataset is another key strategy for improving model performance. A larger and more diverse dataset provides the model with a richer source of information, enabling it to learn more effectively and generalize better to unseen data. By incorporating additional songs with accurate features, researchers can enhance the model’s ability to classify music genres and analyze sentiments accurately. This expansion of the dataset not only improves the performance of existing models but also lays the groundwork for the development of more advanced algorithms in the future. The process of dataset expansion involves collecting and annotating a diverse range of Bengali songs, ensuring that they cover various genres and emotional themes. Careful curation of the dataset is essential to maintain quality and relevance, as well as to minimize biases that may affect model performance. Researchers may leverage crowd-sourcing or collaborate with music experts to gather and annotate the data, ensuring its accuracy and comprehensiveness. Once the expanded dataset is in place, researchers can use it to train and evaluate improved models for genre classification and sentiment analysis. By incorporating a larger and more diverse set of examples, these models can learn more robust representations of Bengali music, leading to better performance on real-world tasks. Additionally, researchers can employ advanced techniques such as transfer learning and ensemble methods to further boost model performance and robustness. Overall, the classification and categorization of Bangla music genres and sentiments are essential for the development of effective online platforms that cater to the diverse preferences of Bengali music enthusiasts. Through iterative refinement and expansion of datasets, researchers can continue to improve the accuracy and efficiency of genre classification and sentiment analysis algorithms, ultimately enhancing the user experience and promoting the discovery of new and exciting music.
PACE: Python AI Companion for Enhanced Engagement
(IUB, 2024-11) Shochcho, Muhtasim Ibteda; Rahman, Mohammad Ashfaq Ur
This research introduces PACE (Python AI Companion for Enhanced Engagement), an AI-driven, interactive tutoring system designed to support Python programming students through personalized, responsive guidance. Addressing common challenges in traditional educational settings, such as limited individualized feedback and scalability, PACE leverages the capabilities of Large Language Models (LLMs) to create a tailored, human-like tutoring experience. Fine-tuned using the Gemma 2B model with scaffolding and conversational datasets, PACE offers step-by-step instruction and adaptive problem-solving support that encourages active learning without directly revealing answers. The development of PACE involved a robust design and implementation process, including backend development with FastAPI, frontend construction in React, and data management in MySQL, to deliver seamless, real-time interactions with students. Interaction design and user interface principles were carefully applied to create an intuitive environment that promotes engagement and reduces cognitive load. The model’s fine-tuning was achieved using Low-Rank Adaptation (LoRA), optimizing computational efficiency while enhancing PACE’s ability to deliver domain-specific instruction in Python. The system was evaluated with undergraduate students using five pedagogical metrics: Relevancy, Correctness, Completeness, Motivation, and Clarity, and included a System Usability Scale (SUS) assessment. Findings demonstrate PACE’s success in fostering understanding, boosting student engagement, and encouraging critical thinking. By tracking student progress and providing scalable, personalized support, PACE also eases the instructional burden on educators in large classes. This work showcases the potential of LLM-based intelligent tutoring systems to deliver cost-effective, high-quality, personalized learning experiences, making them viable for broader educational deployment.
Cost Effective IoT-Based Smart System for Avoidance of Obstacle and Intruder Detection
(IUB, 2024-12) Amrit, Imam Tajnoor Hossain; Talukdar, Md. Shadman Sakib; Takbir, Mahiyan Rahman
This study introduces a cost-effective integrated system that significantly enhances the safety and security of autonomous vehicles by merging obstacle avoidance and intrusion detection functionalities through Internet of Things (IoT) technologies. The system employs ultrasonic sensors for accurate obstacle detection, with an Arduino Uno and Raspberry Pi 3 handling data processing and control tasks. Facial recognition technology is incorporated to monitor and identify unauthorized individuals in real time. This integration allows the vehicle to autonomously navigate while also securing itself against potential intruders, all achieved with readily available and affordable components. The system demonstrates strong performance in detecting and avoiding obstacles in straight forward scenarios, although accuracy may diminish in more complex environments. The facial recognition component achieves consistent and reliable results under controlled conditions. By combining these technologies in a budget-friendly solution, this research offers practical advancements in intelligent and secure autonomous vehicle systems, suitable for applications in various domains including robotics and smart vehicle technology.
GorbhoShongi: a mHealth App for the Mental Well-being of Bangladeshi Pregnant and Postpartum Women
(IUB, 2024-12) Mohtasim, Syed Niaz; Arpita, Faiza Omar; Ahmed, Istiaq
Mental well-being during pregnancy and postpartum periods is significantly overlooked worldwide, especially in developing countries like Bangladesh. It is crucial to look after the mental health of pregnant and postpartum women, as it can have long-lasting effects on both mother and child. We conducted a systematic literature review using the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) method. From 187 initial articles, we carefully analyzed 18 relevant studies about mobile health (mHealth) solutions for maternal mental well-being. This review helped us understand the existing approaches and identify important gaps in current solutions. Based on our literature review, we developed a conceptual design framework for “GorbhoShongi,” a mobile health app specifically designed for pregnant and postpartum women in Bangladesh. We chose features that would be most helpful and relevant to our target users. The app is developed in both English and Bangla to reach a wide audience and ensure accessibility. Our app includes some unique features that set it apart from existing solutions. We added a large language model (LLM) chatbot and daily well-being tips to provide a more personalized and supportive experience. These features aim to offer encouragement and immediate support to users during their pregnancy and postpartum journey. We created a low-fidelity prototype of “GorbhoShongi” to demonstrate the app’s core functionalities and design principles. This prototype serves as proof of concept, show- casing how the app can provide comprehensive mental health support tailored to the Bangladeshi context. The next steps include developing a high-fidelity version of the app using JavaScript frameworks like Next.js and Capacitor.js. We also plan to conduct user testing to evaluate the app’s usability and user experience. In conclusion, “GorbhoShongi” represents an innovative approach to addressing maternal mental health challenges in a developing country. By providing a culturally sensitive, accessible, and supportive mobile health solution, we hope to contribute to improved mental well-being for pregnant and postpartum women and encourage further research in this important field.
Design and Development of mHealth App for Healthcare Professionals’ Stress Management in Bangladesh
(IUB, 2024-12) Rashid, Sami; Rafid, Lishan; Badrul, Tasnuba
Healthcare professionals (HCPs) in lower- and middle-income countries (LMICs) like Bangladesh often face high levels of workplace stress, which can negatively impact their mental well-being. However, despite its significant impact, this issue is frequently over- looked, leaving many HCPs vulnerable to burnout and other mental health challenges. Mobile health (mHealth) tools, integrated with wearable devices like smartwatches, have the potential to play a significant role in managing workplace stress among HCPs by lever- aging physiological signals. This study explored the design and usability evaluation of a user-centered mHealth tool named ‘FreedHCP,’ aimed at managing stress among HCPs in Bangladesh. The research objectives include conducting design requirements and needfind- ing analysis, developing a high-fidelity prototype, evaluating the usability of FreedHCP, and proposing a deep learning (DL) based method for automatic stress detection that could be implemented in future applications. A survey involving 71 HCPs revealed that high workload, patient and family pressure, and staffing shortages were major stressors, while social support, taking short breaks, and time management were effective coping strategies. Participants also indicated a strong preference for app features such as guided meditation sessions, personalized stress management plans, real-time health tracking, and willingness to use a smartwatch-based mHealth app for real-time stress monitoring. Based on these insights, the FreedHCP was developed with five core feature categories: ‘Assigned Tasks’, ‘Notification Settings’, ‘Get Help’, ‘Wellness Check’, and ‘Supervisor Dashboard’. The usability evaluation revealed that the ‘Smart Monitoring’ feature from the ‘Wellness Check’ category was the most liked, with 29.2% of votes. The app achieved a mean System Usability Scale (SUS) score of 71.77. Moreover, an overwhelming 95.8% of participants expressed willingness to use and recommend FreedHCP to colleagues. In parallel, medical students are another vulnerable group significantly affected by mental health challenges such as anxiety, depression, and burnout due to the intense pressures of their academic and clinical environments. These mental health issues, further worsened by a demanding workload and the emotional burden of patient care, can adversely impact both personal wellness and professional development. To address these concerns, a dataset of 886 Swiss medical students was analyzed to automate the screening process for anxiety, depression, and burnout using Machine Learning (ML) and Deep Learning (DL) approaches. The analysis compares the performance of two advanced computational models: an Ensem- ble classifier, integrating Random Forest (RF), Naive Bayes (NB), and Light Gradient- Boosting Machine (LightGBM), and a Deep Neural Network (DNN). The DNN model emerges as the better performing method by demonstrating accuracy rates of 81.4% for depression, 76.65% for anxiety, and 73.59% for burnout. Comparative analyses further validate the DNN’s efficacy against the Ensemble classifier, thereby providing a promising method for automated clinical diagnosis for mental health professionals. This research not only demonstrates the utility of mHealth solutions in stress management but also highlights the promise of Artificial Intelligence (AI) driven models for advancing mental health care in professional healthcare environments.
Optimizing UAV Performance using ArUco Marker-Enabled Vision-based Precision Landing and Autonomous Charging
(IUB, 2024-12) Islam, Tabriji; Salma, Khandaker Umme; Hossen, Ishtiaq
Unmanned Aerial Vehicles (UAVs) have become indispensable in applications such as precision agriculture and environmental monitoring. Despite their potential, their widespread deployment is hindered by three critical challenges: limited flight time due to battery constraints, declining navigation accuracy at higher altitudes, and reliance on Global Positioning System (GPS) signals, which degrade with altitude, environmental factors, and battery depletion. To address these issues, this study proposes a system integrating vision-based ArUco marker detection with a proportional integral derivative (PID) controller for precision landing, paving the way for autonomous recharge and extending operational capabilities. Our approach leverages a combination of GPS-based navigation and vision-based marker detection to ensure accurate landings at target waypoints. At lower altitudes (10 meters), GPS accuracy declines with deviations averaging 0.15 meters, increasing to 0.41 meters at 25 meters. In addition, battery depletion worsens positional inaccuracies, with up to a 152.3% increase in GPS-based landing deviations as the voltage drops from 12.6 to 9.1 V. Vision-based guidance, enabled by ArUco markers, compensates for these inaccuracies, achieving a detection rate of 83.33% at 4.5 meters but decreasing to 33.33% at 6.0 meters. The Vision-guided landing system with PID controller further refines the descent by dynamically adjusting the position of the UAV, reducing the deviation by 56.5% in simulation tests, demonstrating its effectiveness in enhancing the landing accuracy over the GPS-only system. Experimental results validate the effectiveness of the system, showcasing its potential to allow UAVs to autonomously locate charging stations, extend mission durations, and maintain consistent operational performance under varying environmental and battery conditions.
Comparative Analysis of Deep Learning Models for Intracerebral Hemorrhage Detection from Computed Tomography
(IUB, 2024-12) Ahad, Taiseer Rakiin; Rakiin, Taiseer; Nur, Md. Ashaduzzaman
Intracerebral hemorrhage (ICH) is a serious medical condition caused by bleeding within the brain tissue, which can lead to life-threatening complications if not diagnosed quickly. This study compares the performance of three advanced deep learning models, InceptionV3, DenseNet201, and EfficientNetB4, for detecting ICH from computed tomography (CT) scans. This work's novelty lies in preprocessing techniques such as windowing with Hounsfield Units and Skull segmentation to provide cleaner input data and Grad-CAM visualizations to ensure the model’s predictions are interpretable. Using comprehensive evaluations, including metrics like AUROC, precision, and F1 score, EfficientNetB4 achieved the best results with an accuracy of 0.85, precision of 0.87, and F1-score of 0.86, demonstrating its potential to assist medical professionals in making faster and more reliable diagnoses.
An Undergraduate Project on Topic “Design and Development of a Low-cost Bangla Voice Interactive Children Educational Robot “TINY” for ASD Application”
(IUB, 2024-12) Jobaida, Nohori; Mahbub, Ahsan; Khan, Jafrin
Autism Spectrum Disorder (ASD) presents challenges in areas such as communication, social interaction, and learning. To address these challenges, TINY has been developed as an interactive educational robot tailored to meet the unique needs of children with ASD. TINY incorporates adaptive AI technology, allowing it to personalize teaching methods and activities based on each child’s individual preferences and developmental progress. Its features include sensory-sensitive design, engaging interactive elements, real-time emotion detection, and gamified learning to foster cognitive and social skills in a stress-free environment. The robot also provides a valuable link between caregivers and professionals by generating detailed progress reports and actionable insights. These reports help in understanding the child’s learning patterns and informing targeted therapeutic approaches. By combining advanced technology with empathy-driven design, TINY aims to create a nurturing educational experience that promotes skill development, enhances emotional well-being, and supports overall growth in children with ASD. This study outlines the design, functionality, and impact of TINY, demonstrating its potential to revolutionize learning and therapy for children with ASD.
Dual-Task Real-Time Low-Light Lane and Pothole Detection for Resource-Constrained Environments
(IUB, 2025) Alam, Md Iftekharul
Lane detection and road hazard awareness are crucial for ensuring safety in autonomous driving and Advanced Driver-Assistance Systems (ADAS). These systems rely heavily on clear visual cues, which are often compromised in low- light driving scenarios. The challenge is especially pronounced in low- and middle-income countries (LMICs), where poorly illuminated roads, faded lane markings, and unmaintained sur- faces frequently co-occur. Under such conditions, conventional single-model detectors trained for daytime environments degrade sharply, as lane cues and pothole textures often compete in the same field of view. To address this, we present a lightweight dual- model pipeline that integrates a low-light enhancement front end with an OpenCV-based lane delineation pipeline and a YOLOv12 detector for pothole localization. The models run in parallel on shared inputs, and their outputs are fused to generate a unified lane geometry and hazard map in a single pass. The architecture is optimized for modest compute and memory budgets, enabling deployment in resource-constrained settings while maintaining high throughput. Evaluated on evening-time urban road scenes from Bangladesh, achieves 88.7potholes and 89.3FPS on NVIDIA GTX 1050Ti, outperforming a single-detector baseline. These results highlight the potential of our approach for practical, real-time ADAS perception in underserved regions. Index Terms—Low-light imaging, Lane detection, Pothole de- tection, YOLOv12, OpenCV, Image enhancement, Edge computing Autonomous driving
U flow-net: integration of flow estimation and kernel-based method in a modified u-net architecture for video frame interpolation
(IUB, 2025-01) Tabib, Fatin Israq; Khatun, Halima; Sarker, Shuvro Sourav
architecture relies on the addition of flow estimation layers during the downsampling process, which allows relevant motion details to permeate into frame synthesis. This design, in conjunction with a progressive refinement approach, guarantees high-fidelity output while remaining computationally efficient. Extensive experimental evaluations with Tarzan were performed using the Vimeo90K, UCF101, and Middlebury datasets. Extensive experiments evaluate UFlow-Net's performance against many previous optical flow models in terms of Peak Signal-to-Noise Ratio (PSNR) and Structural Similarity Index (SSIM) of the interpolated results, with improved maintaining of structural detail and spatial coherence in UFlow-Net interpolated frames. Though UFlow-Net outperforms current methods by a large margin, it may still struggle in very complex background scenes or images with highly rapid motion. The following are limitations that guide further studies in video frame interpolation technology. This research has significant practical implications. The accuracy and computational efficiency balance of UFlow-Net renders it especially applicable to real-world video enhancement tasks. This work makes a major step forward in the development of high-quality video processing systems and looks positive for future work.
Blockchain based Voluntary Carbon Market for Bangladesh
(IUB, 2025-04) Mahmud, Mohammad Aftab; Mahmud, Md. Ridwan
Bangladesh is identified among the countries most vulnerable to climate change, facing significant environmental threats, including deforestation, rising carbon emissions, and severe climate impacts. Despite the critical need to mitigate these environmental challenges, Bangladesh currently lacks a structured and efficient voluntary carbon credit market. This absence hinders the potential of businesses, organizations, and individuals from effectively offsetting their carbon footprints and participating in sustainable environmental practices. A transparent, secure, and regulated carbon credit trading platform is urgently required to encourage widespread adoption of sustainability initiatives. The primary objective of this project is to design and implement a decentralized blockchain-based marketplace specifically adapted for voluntary carbon credit trading in Bangladesh. The platform aims to provide a secure, transparent, and user-friendly environment for businesses and individuals to engage in carbon offsetting activities. Additional objectives include establishing a clear incentive model that encourages active participation and laying foundational frameworks for future integration with environmental regulatory bodies in Bangladesh. Leveraging blockchain technology, this project proposes a robust, decentralized infrastructure designed to significantly enhance transparency, immutability, and security within carbon credit trading. Smart contracts form the backbone of the platform, enabling automated and secure peer-to-peer transactions without intermediaries, thus reducing opportunities for fraud or mismanagement. The system includes an intuitive, web-based frontend interface built using Next.js, allowing seamless access for users wishing to buy, sell, or verify carbon credits. Furthermore, integrated data analytics functionalities provide comprehensive insights into carbon credit activities, enabling continuous monitoring, reporting, and informed decision-making. This project successfully developed a scalable prototype platform demonstrating how carbon credits can be transparently and efficiently traded in Bangladesh. Core functionalities of the Minimum Viable Product (MVP) include secure token issuance representing carbon credits, decentralized peer-to-peer trading, robust credit verification mechanisms, and transparent blockchain- based transaction logging. The platform incorporates wallet authentication using Meta- Mask, enhancing ease of use and security for participants. The established MVP clearly demonstrates the feasibility of blockchain-based carbon markets in Bangladesh and sets a foundational path towards regulatory compliance, structured governance, and the development of sustainability-driven incentive models. Overall, this initiative addresses critical gaps within Bangladesh’s environmental landscape, offering a pioneering approach to voluntary carbon credit trading. The platform’s decentralized nature ensures transparency, fosters trust among stakeholders, and establishes groundwork for seamless future integration with national and international environmental frameworks.