A Feature Fusion-Based Lightweight Algorithm for Small Object Detection in UAV Aerial Imagery

To address the issue of low detection accuracy for small objects in drone aerial imagery, this paper improves the YOLOv8n framework. The backbone network is replaced with an improved EfficientNetV2, and a Multi-Scale Feature Aggreg-ation Pyramid is introduced to enhance semantic understanding of features. Validation on the VisDrone-2019 dataset demons-trates that the proposed improvements significantly increase the mAP, proving that the model’s ability to detect small objects in aerial images has been enhanced.