Hybrid CNN-transformer model-based object detection methodology for UAV imagery

Abstract

This paper proposes a hybrid object detection method for UAV imagery by combining YOLO (CNN-based) and RT-DETR (transformer-based) models. The framework integrates a bounding box processing module with fusion strategies and a confidence-based selection mechanism. Experimental results show that the hybrid approach improves detection accuracy, achieving a 0.7 increase in mAP50.