MFS-YOLO: An Object Detection Network for Drones

UAV aerial photography has been widely used in many fields. There are some problems in UAV aerial image object detection, for instance the small target and high aspect ratio target aggregation, and low detection accuracy. So we propose MFSYOLO. First, CMSC is fused into the C3K2, then Introduce SDI in the neck, P2 detection is added, and WIMIoU loss function is adopted. The paper conducted experiments on VisDrone dataset. Compared with the benchmark network, the indexes of mAP0.5 and <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$m A P 0.5: 0.95$</tex> of MFS-YOLO are 0.443 and 0.269. It is 0.05 and 0.031 higher than YOLO11s. The results show that the improvement is effective.