https://repository.sustech.edu/handle/123456789/9171 2026-09-01T15:38:38Z https://repository.sustech.edu/handle/123456789/28498 Title: Performance Analysis of OFDMA System in Multi-User Environments Authors: Othman, AbdulRahman Tawfig Muhammad; Abdalla, Ayoub Osman Ibrahim; DaffaAllah, Ibrahim Mohammed Babiker; Salih, Kamal Ali Alzubair; Ali, Obada Idris Abas Abstract: This project presents a comprehensive performance analysis of an Orthogonal Frequency Division Multiple Access (OFDMA) system in multi-user environments. The study investigates the impact of user mobility, varying signal-to-noise ratios (SNR), and different channel conditions on key performance metrics, including Bit Error Rate (BER), throughput, and spectral efficiency. The analysis differentiates between static and mobile users, highlighting how mobility introduces challenges such as the Doppler Effect and increased interference, which lead to performance degradation. In static scenarios, users experience a stable channel environment that results in lower BER and higher throughput as SNR improves. However, in mobile scenarios, the BER tends to be higher, particularly at lower SNR values, due to the unpredictable nature of signal fluctuations caused by user movement. This variability adversely affects throughput and spectral efficiency, making it crucial to implement effective resource allocation and management strategies. The findings underscore the importance of optimizing system performance in diverse user conditions, emphasizing that while static users benefit from consistent channel quality, mobile users require adaptive techniques to mitigate the adverse effects of mobility. Overall, this research contributes valuable insights into enhancing OFDMA system performance in real-world applications, ensuring efficient communication in both static and dynamic environments. 2025-01-01T00:00:00Z https://repository.sustech.edu/handle/123456789/28497 Title: Crop Health Monitoring Using Convolutional Neural Network (CNN) Authors: Othman, Abdalkareem Abdalmajed Mohmed; Hag Elzaki, Khalid Elwalid Omer; Babiker, Mohammed Osama; Ali, Omer Ibrahim; Supervisor, Rashid Saeed Abstract: Crop diseases threaten global food security, requiring rapid and accurate detection to maintain agricultural productivity. Traditional methods for monitoring crop health are costly and inefficient. This research explores the use of convolutional neural network (CNN), a type of deep learning algorithm, to automate plant disease detection and classification through image analysis. Leveraging publicly available datasets of plant leaf images, a robust model is developed and trained to recognize different stages of crop diseases. The CNN model is improved using techniques such as data augmentation and evaluated through metrics such as precision, accuracy, recall, and F1 score. The results demonstrate the potential of CNN to improve the accuracy and effectiveness of disease detection, contributing to sustainable agriculture and improved crop management practices. Limitations and future work include expanding the dataset, improving real-time data processing, using CNN fast Fourier transform, increasing the used layers for training, and generalizing the model to a wider range of crops and diseases 2025-02-01T00:00:00Z https://repository.sustech.edu/handle/123456789/28496 Title: Real-Time Object Detection and Tracking for Search and Rescue Operations Authors: Aber, Ahmed Mohamed Salih; Abdalla, Mohamed Abdalla Mohamed; Alkhuzaei, Muaadh Ali Saeed; Omer, Omer Mohamed Abdalla; Supervisor, Rashid A. Saeed Abstract: Real time object detection and tracking are of significant importance in improving search and rescue (SAR) by offering the ability to quickly and efficiently identify goals in adverse environments. These systems build on developments in computer vision and deep learning and are comprised of superordinate algorithms such as CNNs and transformers to process visual data from drones or cameras and other external sensors. This element guarantees the fast localization of objects such as humans, debris, or vehicles and constant tracking of those in motion, even at crowded scenes. Some of the main issues treated involve low illumination environments; occlusions; and, the practical requirements of real-time implementations compatible with the power-constrained environment of edge hardware. With the help of certain methodologies used in these systems such as multi-modal sensor fusion and adaptive tracking the chances of success for SAR are enhanced simply because these systems sharpen the techniques used in the search and rescue missions besides minimizing the response time. To this end, the methodologies, tools, and the possible use of real-time object detection and tracking technologies for SAR contexts are discussed in this paper with an aim of demonstrating how they enhance the desirable changes to disaster response operations. 2024-01-01T00:00:00Z https://repository.sustech.edu/handle/123456789/28495 Title: Fire Detection based Image Recognition Using Machine Learning Authors: Suliman, Alfatih Abdelbagi Adam; Adam, Eissa Khmeas Eissa; Wadi, Ismail Shaibo Abdalla; Hamid, Mohamed Hamid Hary; Adam, Mohammedalgasem Eltalib Mohammed; Supervised, Rashid A Saeed Abstract: Fire detection is vital for safeguarding both the environment and human life, however, conventional fire detection methods, such as temperature and smoke sensors, often face inherent limitations. The machine learning models, specifically convolutional neural networks (CNNs), are commonly used for processing images to enhance the accuracy and efficiency of fire detection, enabling these systems to detect fire in real-time. We propose the YOLOv8 (You Only Look One Version8) model it’s capable of analyzing high-resolution images of both fires and non-fire scenarios from diverse sources. The Fire dataset is typically split into 80% training, 15% validation, and 5% testing to ensure the model learns effectively. The accuracy of bounding box predictions during training the box loss starts at around 0.54 and progressively decreases to about 0.22, and the Recall starts at around 0.94 and steadily increases to about 0.99, the loss decreases from about 0.88 to 0.76, showing that the model is making fewer errors in bounding box regression on the validation data, a significant decrease in loss values, suggesting effective learning and optimization, while high and improving precision and recall demonstrate its strong detection performance. The consistently increasing mAP (mean Average Precision) values further indicate that the model is accurately predicting bounding boxes and classifying fire objects. Future work will focus on improving data, models, and integrating this approach with systems like firefighting control. 2024-10-01T00:00:00Z