International Science and Technology Journal

This study aims to train and test a deep learning model using artificial intelligence techniques, specifically You Only Look Once version 4 (YOLOv4), for real-time automatic detection of building extension violations from Drone images in Al-Ajaweed district )Jeddah city(. The model was trained on a dataset of 800 images, and cloud computing resources from Google Colab were utilized. After training, the model was applied to Drone images to detect violations in the study area, where 256 samples were monitored. Among these samples, 67 buildings had violations, with the extension area exceeding 50%, while 189 buildings had no violations.The model showed high accuracy in predicting violations at 91% and 87%, as well as high recall accuracy at 98% and 93%, with an F1 score of 0.95 and 0.90. It also demonstrated the ability to detect spatial overlap at 84% and 79% in the training and test datasets, respectively. Furthermore, it showed an average geometric accuracy of 7.1%.The study emphasized the significance of modern technologies in supporting building violation monitoring and highlighted the potential of the trained YOLOv4 model for real-time detection of violations and non-compliance with building regulations. Keywords: Automatic detection, building violations, real-time, aerial Drone images, artificial intelligence, YOLOv4, Jeddah city.