Detection of Airports and Runways From Aerial Imagery Using Deep Learning

Disaster management calls for as quick a detection of critical infrastructure-like airports and runways. Most of the traditional object detection models, which include CNNs, R-CNN s, SSDs, and earlier versions of YOLO, have high computational costs and low accuracy, particularly for smaller objects within aerial and satellite imagery. It presents the YOLOv9 object detection model with very significant improvements in feature extraction, bounding box prediction, and several modern techniques involving attention mechanisms and multi-scale feature fusion. For the purpose of improving model robustness, extensive data augmentation was adopted in training. The proposed model was tested and validated on the diverse dataset of aerial and satellite images, attaining an accuracy as high as 91.3% with a low false alarm rate of just 7.08%. These further improvements will make YOLOv9 highly scalable to support disaster management with faster infrastructure assessments, improvement in border surveillance, among others.