A lightweight multi-scale framework with edge enhancement and attention for small object detection in UAV imagery
Small object detection in unmanned aerial vehicle (UAV) images is challenging due to small object scale, dense distribution, and complex background interference. These factors make it difficult to balance detection accuracy and computational efficiency. To address this issue, this study proposes a lightweight real-time object detection framework, termed EAMS-YOLO. The method reconstructs the detection scale by introducing a high-resolution P 2 feature layer and removing the P 5 feature layer. This design enhances fine-grained feature representation for small objects. Meanwhile, a multi-scale edge information enhancement module and a multi-scale linear attention module are designed to strengthen fine-grained feature representation and cross-scale contextual modeling. In addition, a Shape-IoU-based shape-aware loss is introduced to improve object localization accuracy. Experiments on the VisDrone2019 dataset show that, compared with the YOLOv11 baseline, EAMS-YOLO achieves gains of 6.2% and 3.4% in mAP50 and mAP50:95, respectively. Meanwhile, the number of model parameters is reduced to 0.9 M. Experimental results on the DOTA and PASCAL VOC datasets indicate that the proposed method provides stable performance gains across different scenarios and object scales. Furthermore, in engineering degradation tests and edge device evaluations, the model maintains robust detection performance under challenging imaging conditions and resource-limited environments. The proposed EAMS-YOLO provides a balanced trade-off among detection accuracy, model complexity, and deployment adaptability, making it well suited for UAV-based small object detection in resource-constrained environments.