A Novel Feature Extraction Network for YOLOv11: Enhancing Small Object Detection Performance

Due to the low resolution, complex background and limited pixel data of aerial images, the detection of small objects in aerial images has met great challenges. In order to improve the detection accuracy, our improved feature extraction module is added to the YOLOv11 backbone network. The method adopted in this paper improves the convolution operation. This captures fine-grained details and also improves the presentation of small objects in complex environments. Experiments carried out on the VisDrone2019 dataset show some phenomena, which indicate that compared with the baseline YOLOv11, mAP50 of the proposed method has increased by 0.9%, with only a slight decrease in computational efficiency. The improved model shows excellent capability in identifying small objects and blocking objects. It has proved its value in applications such as aerial surveillance, traffic monitoring, and urban planning.