Rema Albarakati

تخصص الذكاء الاصطناعي · دفعة 2025

Energy Transition During Energy Crisis

This project addresses the energy crisis in Cape Town, South Africa, where frequent power outages and infrastructure failures have led to inequitable energy access. Wealthier residents can afford solar installations, while low-income communities are left behind. My team and I aim to develop deep learning models to detect and classify different types of solar energy systems (solar panels, solar water heaters, and pool heaters) using high-resolution satellite imagery. The project utilizes image segmentation (U-Net) and object detection (YOLO) models to identify and locate solar installations across the city. With a dataset comprising over 2,800 satellite images and nearly 20,000 annotated solar objects, my team and I are working to improve model accuracy, reduce false positives, and enable multiclass classification.

المشرف: Biz Yoder

اطّلع عليه في تقرير البرنامج (PDF)

تخصص الذكاء الاصطناعي · دفعة 2024

Deep Learning-Based Fault Detection System for Structural Defects in Nano-Quadcopters

Detecting faults in nano-quadcopters is critical to ensuring safety and performance. This study proposes a deep learning-based system for detecting structural defects in real time. Utilizing advanced feature engineering and dimensionality reduction techniques, the proposed model offers a reliable and efficient solution for ensuring the integrity and functionality of nano-quadcopters in various applications with high accuracy and low inference time.

اقرأ التقرير النهائي (PDF)

اطّلع عليه في تقرير البرنامج (PDF)

تخصص الذكاء الاصطناعي · دفعة 2023

Detecting Inappropriate Content in Short Videos for Children

The rise of user-generated content on platforms like TikTok and YouTube Shorts has raised concerns about inappropriate content reaching young audiences. These platforms rely on AI-driven recommendation systems to show tailored content. However, these systems might unintentionally suggest unsuitable material. Our project proposes a solution that combines the InceptionV3 CNN-RNN model for video analysis and the BERT model for understanding text context. The goal is to accurately identify and address inappropriate content in videos and their text descriptions. The dataset consists of 2,255 videos from YouTube Shorts and TikTok. After trying various approaches, the modified InceptionV3 model showed the best results, achieving an accuracy of 84.48%. This is a significant improvement over previous methods.

اطّلع عليه في تقرير البرنامج (PDF)