PPA-enhanced YOLOv10: real-time detection of small UAVs in complex environments

Unmanned Aerial Vehicles are being used more and more for surveillance, traffic monitoring, and disaster response in urban environments. This causes big challenges for urban security and privacy. It's still really hard to detect small drones in the messy and changing urban environment because of things like unclear object boundaries, environmental changes, and background distractions. To deal with these challenges, this study suggests a new anti-drone detection framework. It makes the YOLOv10 model better with a Pyramid Pyramid Attention (PPA) module. This improvement mixes multi-branch feature extraction, attention mechanisms, and anchor size optimization through k-means clustering. It's especially made for detecting small drones. The experiment results using the TIB-Net drone dataset show that the PPA-enhanced YOLOv10 model is a lot better than the baseline YOLOv10. It improves mAP50 by 4.1%, mAP50-95 by 0.9%, precision by 4.9%, and recall by 1.1%. Even though the computing cost of this method goes up a little bit, it gives a hopeful solution for real-time detection of small drones, especially in complex urban environments.