Multi-size drone detection using YOLOv5 network

Abstract

With increasingly modern technology, advanced and flexible functions, compact design, and low cost, drones are recently used in many fields with different effective purposes. Unlike beneficial applications, hostile forces leverage drones to explore the terrain, carry illegal explosives, and so on. Those applications can seriously threaten national security and defense. To prevent illegal drones effectively, we apply deep neural networks to detect the illegal drones in a variety of conditions and different sizes of drones. Accordingly, a computer-based system using modern cameras combined with an algorithmic model can solve the complex drone detection problem. Therefore, an emerging complex neural network approach based on YOLOv5 is proposed in this paper. With the method, we achieve a very expected result (confidence of 0.993 @0.5IOU), which meets the requirements of the drone detection problem.