Fusion of Deep Cross Block Stage and YOLOv9 for Enhanced Small Object Detection in Aerial Imagery

Small object detection is used extensively in real life and is a key component of computer vision. However, because of their poor resolution and ambiguous features, small objects are frequently missed, lowering the detection accuracy. To tackle the problem, this paper presents a deep learning framework for small object detection in aerial imagery, focusing on small object detection challenges with major emphasis on vehicle detection. The proposed framework integrates Cross Block Stage modules with YOLOv9, enhancing feature representation and multi-scale processing capabilities. Experimental evaluation on the VEDAI dataset demonstrates significant performance improvements, achieving 0.873 accuracy compared to current architectures including FFCA-YOLO (0.748), YOLOv5m (0.723), and YOLOv8m (0.686). The framework exhibits strong category-specific performance, with car and truck detection accuracies of 0.926 and 0.820 respectively, establishing its effectiveness for aerial vehicle detection applications.