http://ar.cou.ac.bd:8080/xmlui/handle/123456789/23 2026-05-28T18:12:49Z http://ar.cou.ac.bd:8080/xmlui/handle/123456789/147 ROAD LANE LINE DETECTION USING U-NET ARCHITECTURE FOR SELF DRIVING CARS Akter, Mosammat Sania; Mouri, Ifrat Jahan This project introduces a deep learning-based road lane line detection system using the U-Net architecture, aimed at enhancing autonomous driving and advanced driver assistance systems (ADAS). The model is trained on the CULane dataset, leveraging semantic segmentation to accurately detect lane markings in diverse road conditions. Preprocessing techniques such as resizing, normalization, and data augmentation are applied to improve model robustness. The system employs binary cross-entropy loss and is optimized using the Adam optimizer for efficient learning. A custom data pipeline ensures smooth training and evaluation, with real-time predictions visualized for performance assessment. The trained model effectively identifies lane boundaries, even under challenging lighting and occlusion scenarios. Extensive experiments validate the system’s accuracy and reliability in real-world conditions. The model achieves an accuracy of 96.15%, outperforming many existing lane detections models. The proposed approach contributes to safer and more intelligent vehicle navigation by reducing lane departure risks. Future enhancements include integrating temporal consistency for improved detection stability. 2025-02-02T00:00:00Z http://ar.cou.ac.bd:8080/xmlui/handle/123456789/129 TRAFFIC SIGN DETECTION UNDER ADVERSE CONDITIONS Mohiuddin, Mohammad; Hasan, Mehedi Automatic traffic sign detection and recognition play a crucial role in real-world applications such as autonomous driving systems and driver assistance technologies. This project aims to develop a system using deep learning techniques to enhance road safety and assist autonomous vehicles by automatically detecting road signs and signals in adverse situations such as intense lighting, snowfall, and rainy conditions. In this project, we utilize a Convolutional Neural Network (CNN) algorithm to train the model using a large dataset provided by the German Traffic Sign Recognition Benchmark (GTSRB). Additionally, OpenCV is employed for image processing, while TensorFlow/Keras is used for model training and image classification. The proposed model achieves an accuracy of 95.33% on the test dataset, demonstrating its robustness in detecting traffic signs under challenging conditions such as intense lighting, foggy weather, or heavy snowfall. This system can be integrated into driver assistance systems and autonomous vehicle. In the future, it can be further improved for real-time traffic sign detection and enhanced to achieve higher accuracy in complex environment 2025-02-01T00:00:00Z http://ar.cou.ac.bd:8080/xmlui/handle/123456789/128 SYMPTOM-BASED DISEASE PREDICTION AND PRECAUTION SYSTEM USING MACHINE LEARNING & DEEP LEARNING Hasan , Md Fuad The joining of machine learning and common dialect handling (NLP) has changed healthcare, empowering mechanized frameworks for foreseeing and diagnosing maladies. This consideration pointed to create an progressed system for malady classification utilizing literary portrayals of indications. The system's establishment may be a K-nearest neighbor (KNN) classifier, chosen for its direct approach and demonstrated viability in classification errands. Side effect portrayals are preprocessed utilizing TF-IDF vectorization, changing over content into numerical information reasonable for machine learning calculations. To progress information comprehension, exploratory information examination (EDA) strategies, counting word cloud visualization and stop word disposal, are utilized. The inquiry examines modern Neural Arrange structures, such as CNN, Bi-LSTM, GRU, LSTM, CNN+GRU, RNN, and CNN+LSTM, nearby routine machine learning models like Back Vector Machine (SVM), Calculated Relapse (LR), Choice Tree (DT), Multinomial Gullible Bayes (MNB), and Irregular Timberland Classifier (RFC). Besides, transformer-based models, counting BERT and DISTILBERT, are utilized to tackle state-of-the-art NLP capabilities. Execution evaluation was conducted utilizing exactness measurements, classification reports, and perplexity frameworks, with the Choice Tree (DT) demonstrate developing as the best entertainer. To improve availability, a Gradio-powered intuitively chatbot was made, empowering clients to input symptoms and get infection forecasts in conjunction with preventive measures. This inquires about underscores the potential of coordination NLP methods with machine learning to upgrade symptomatic accuracy and productivity in healthcare. By combining prescient analytics with user-friendly interfacing, this extend contributes to the extending field of AI-driven restorative arrangements, advertising a versatile and viable device for infection determination and quiet care. Social media stages have advanced into a essential source of news utilization for groups of onlookers. The expansion of deluding data over conventional media sources, counting social organizing locales, news web journal posts, and online gatherings, has made it challenging to recognize valid news sources 2025-02-01T00:00:00Z http://ar.cou.ac.bd:8080/xmlui/handle/123456789/126 Quantum Walk-Enhanced Hybrid Routing: Integrating Local and Non-Local Best-Effort Strategies for Robust Quantum Networks Dhruba, Anjan Das; Hasan , Md. Mehedi network nodes. Unlike classical networks, quantum networks require specialized routing strategie due to the no-cloning theorem, entanglement decay, and limited quantum memory. This paper investigates the performance of different distributed routing algorithms for entanglement distribution in a quantum internet. We analyze five routing algorithms, including modified greedy routing, local best effort, and non-local best effort approaches. Additionally, we propose a hybrid algorithm that combines local best effort and non-local best effort routing to optimize entanglement distribution. To further improve performance, we introduce an enhanced version of the hybrid algorithm that leverages quantum walks for path discovery and selection. Our study evaluates these algorithms through extensive simulations across multiple network topologies, including grid, ring, and hierarchical structures. We also examine the impact of different virtual graph models—deterministic, power-law, and uniform virtual graphs—on entanglement routing efficiency. Our study evaluates these algorithms through extensive simulations on multiple network topologies—grid, ring, and hierarchical structures—as well as different virtual graph models, including deterministic, power-law, and uniform virtual graphs. The results reveal that the hybrid best effort algorithm (d=2) is the most consistent performer, demonstrating low latency, high fidelity, and stable scaling across all network topologies. The quantum-enhanced hybrid approach achieves the highest fidelity in most cases but exhibits occasional fluctuations and slightly higher latency. Meanwhile, the non-local best effort algorithm (d=2) proves to be the best choice for latency-sensitive applications, maintaining low latency and reliable scaling. Our findings suggest that integrating quantum walks into hybrid routing strategies can significantly enhance fidelity without sacrificing scalability, making it a promising approach for future quantum networks. Furthermore, our analysis provides practical insights into selecting optimal routing algorithms based on specific network requirements—whether prioritizing fidelity, latency, or overall stability. This work contributes to the ongoing development of scalable and efficient quantum routing protocols, paving the way for real-world deployment of quantum internet infrastructures. Future research will explore adaptive learning-based routing strategies to further improve the robustness and adaptability of quantum network routing 2025-02-01T00:00:00Z