Lightweight Target Detection for UAVs Based on YOLOv8 and AirSim

In the field of UAV applications, accurate image recognition is crucial for tasks such as reconnaissance and inspection. However, images acquired by UAVs often suffer from low resolution due to their maneuverability and complex environments. Additionally, due to payload capacity limitations, Unmanned Aerial Vehicles (UAVs) face challenges in supporting conventional models due to computational resource constraints. To address these issues, this paper proposes a lightweight UAV target detection model, YOLOv8-spdb, based on YOLOv8, which was successfully deployed in the UAV simulation environment AirSim. The model integrates the downsampling module SPDConv and the dynamic sparse attention module BiFormer, which better adapt to the recognition of small targets in images captured at high speed by UAVs. Simulation experiments demonstrate that compared with other models, YOLOv8-spdb achieves a significant improvement in target recognition accuracy while maintaining only a minimal increase in parameter count.