YOLOv8s-AeroLite: A Lightweight and Aerial-Optimized Detector for UAV-Based Object Detection on the VisDrone Dataset

UAV-based vision systems face major challenges including small object size, dense scenes, occlusions, and limited onboard computation, yet they are crucial for applications such as traffic monitoring, public safety, and urban surveillance. This paper presents YOLOv8s-AeroLite, a lightweight detection framework tailored for UAV-based vehicle monitoring in complex aerial environments. Building on YOLOv8s, the model integrates GhostC2f for efficient feature extraction, CBAM-Lite and AerialLKA for spatial-channel attention enhancement, and a GFLv2 prediction head for quality-aware localization, with GELU activation improving convergence in compact networks. Evaluated on the VisDrone benchmark, YOLOv8s-AeroLite achieves 51.6% $\mathrm{mAP} \text{@} 0.5$ using only 8.2 M parameters and 21.1 GFLOPs, delivering strong small-object performance while maintaining low computational overhead. These results demonstrate its suitability for real-time UAV surveillance and edge deployment.