Lightweight and Efficient Air-to-Air Unmanned Aerial Vehicle Detection Neural Networks

This paper introduces a lightweight approach for detecting distant aerial targets using onboard camera mounted on unmanned aerial vehicle (UAV). Building upon YOLOv8, we propose the integration of the C3Ghost algorithm to enhance the backbone network, reducing model parameters. We also employ the effective feature fusion (EFF) module to achieve more comprehensive feature fusion. Additionally, a novel detection box loss function is proposed. The effectiveness of these improvements is validated on a dataset, demonstrating significant performance gains in the task of detecting small targets.