Deep Learning Techniques for Early Visual Pollution Predic-tion in Facades: High-rise Buildings in Kurdistan vs. Burj Kha-lifa.

In recent years, architectural visual pollution from high-rise buildings has emerged as a pressing issue in developing regions such as Kurdistan-Iraq, negatively impacting urban aesthetics and residents' well-being, yet it remains underexplored in research and practice. The research gap identified was a lack of knowledge about the concept of visual pollution in preoccupancy issues, which related to design itself and its impact on individual facades. This study aims to bridge this gap by developing a practical and theoretical framework for predicting facade-related visual pollution early in the design process by training a deep learning model composed of four YOLOv11 convolutional neural networks, each dedicated to a theoretically derived indicator: identical repetition, absence of individuality, no volume breakdown, and equal size for all floors. The study includes fifteen local high-rise buildings and compares them against an international benchmark, the Burj Khalifa. All facade images were captured using drone photography to guarantee high accuracy. The results indicate a high level of visual pollution, an average of 86%, in local cases, while the Burj Khalifa scored 0% across all indicators. The model confirmed the effectiveness in detecting design flaws contributing to visual pollution, with detection accuracies of 99% for Model 1, 91% for Model 2, 89% for Model 3, and 75% for Model 4. This research demonstrates the potential of deep learning as a predictive tool for identifying visual pollution early in the design process.