Enhancing small object detection in UAV imagery through edge-preserving context fusion

Abstract Unmanned Aerial Vehicles (UAVs) are pivotal in tasks such as traffic monitoring and rescue operations, yet detecting small objects in UAV images remains challenging due to their diminutive size, low resolution, and complex backgrounds. To address this, we propose ECF-YOLO, an enhanced network based on YOLOv11n, featuring three core modules: LoGStem for preserving geometric edges, the Direction-Aware Context Block for enhancing contextual understanding, and the Efficient Path Aggregation Neck for optimizing multi-scale feature fusion. Experimental results on the TinyPerson and VisDrone2019 datasets demonstrate significant improvements in detection performance, with mAP@0.5 increasing from 14.3% to 22.2% on the TinyPerson dataset. This study highlights the importance of edge preservation and contextual modeling in enhancing small object detection in UAV imagery. The source code will be available at https://github.com/YuIO26/ECF-YOLO