Unmanned Aerial Vehicle (UAV) object detection plays a pivotal role in various fields such as civilian, business, and defense, but it faces obstacles such as different target dimensions, a preponderance of smaller targets, and complex environments filled with densely packed objects and blockages. To this end, we optimize the YOLOv8n model and introduce a novel small target detection model called FSAR-YOLO. First, we integrate a DySample-inspired dynamic upsampling technique, DTsample, which improves upsampling’s efficiency and effectiveness by optimizing the dynamic sampling factor and employing dynamic point sampling. We then introduce an LSKA-based attention mechanism, Rlska, which fuses with the spatial group-wise enhancement (SGE) attention mechanisms and significantly improves the detection of the critical details of small targets. Furthermore, we integrate a detection head designed for small targets and propose an adaptive re-parameterised spatial feature fusion technique aimed at enhancing the re-extraction of small targets. This model also uses the FocalerCIoU loss function for bounding box regression to increase the focus on intersecting samples, thereby improving positioning accuracy in complex environments. Tests conduct on the VisDrone2019 dataset and evaluation of its generalization on the PASCAL VOC dataset show notable advances in the detection of small targets, with a 6.8% increase in mAP compared to YOLOv8n and a 43.3% reduction in parameter volume compared to YOLOv8s, while maintaining comparable detection performance. These results confirm the efficiency and significant advances of our novel methodology in small object detection tasks.